NewYour coding agent can read the release notes before it upgrades.Set up the MCP server →
PyPI, #1876 most downloaded on PyPI
Ultralytics YOLO 🚀 for SOTA object detection, instance segmentation, semantic segmentation, depth estimation, classification, pose estimation, oriented object detection, and multi-object tracking.
Latest stable
8.4.210
Last release 8 days ago
10 Nov 2026
Ships on a steady schedule
a new release about every 8 days
Nearly every release is documented
notes for 60 of the last 60 stable releases
11 versions withdrawn
withdrawn after publishing
5 years old
885 releases · first in 2022
Drag across the ruler to compare, or tap a release to open it.
Ultralytics v8.4.73 is a stability-focused release that mainly improves INT8 TensorRT export reliability on RTX GPUs 🛠️⚡, while also fixing a memory i
Ultralytics v8.4.73 is a stability-focused release that mainly improves INT8 TensorRT export reliability on RTX GPUs 🛠️⚡, while also fixing a memory issue in predict(), improving input order handling, and polishing examples and docs.
Major fix: safer INT8 calibration on RTX GPUs 🚀
int8=True, RTX cards using the NvTensorRTRTXExecutionProvider now perform calibration on the CPU instead of cuda:0.Fix for out-of-memory errors when predicting from large file-path lists 💾
model.predict() could load too much into memory and trigger GPU OOM.Prediction input order is now preserved 📂
Better error handling in the OpenCV ONNX example 🖼️
FileNotFoundError if an image path is missing or unreadable.Small documentation and comment cleanups ✍️
QueueManager doc example so copied code uses the correct region format.CI installation step made more resilient 🔁
More dependable INT8 export on RTX hardware ✅
No model accuracy change from the RTX calibration fix 🎯
Better scalability for large inference jobs 📈
predict() on long image lists, memory use should be much more manageable.More predictable automation workflows 🤖
Friendlier debugging and onboarding 🙌
Overall, v8.4.73 is less about new models and more about making export, inference, and examples more robust and predictable—with the standout improvement being INT8 TensorRT export stability on RTX cards ⚡
Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.4.72...v8.4.73
Your coding agent can read these notes before it upgrades. Set up the MCP server →
v8.4.72 is a small but important stability release that mainly fixes a TensorRT INT8 export crash on some RTX GPUs 🚀, while also improving export envi
v8.4.72 is a small but important stability release that mainly fixes a TensorRT INT8 export crash on some RTX GPUs 🚀, while also improving export environment reliability and cleaning up docs/CI.
Fixed TensorRT INT8 export crashes on certain RTX cards 🔧
format="engine", int8=True could fail on some RTX GPUs, including cases where ONNX Runtime exposed both TensorRT execution providers.Improved TensorRT/ONNX export image compatibility for CUDA 12 🐳
onnxruntime-gpu to below 1.27.0 in the export Docker image.GitHub Actions checkout updated from v6 to v7 ⚙️
actions/checkout version.Documentation updates and cleanup 📚
More reliable INT8 TensorRT exports on RTX hardware 💡
Better support for modern RTX environments 🖥️
Safer Docker-based export workflows 🛡️
onnxruntime-gpu prevents version mismatches that could break ONNX inference and export tests inside CUDA 12 environments.No major model architecture changes 📌
Practical benefit for users ✅
Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.4.71...v8.4.72
Ultralytics v8.4.71 is mainly a release-tag update 📦, but the most important underlying change is a major refactor of the official C++ examples that m
Ultralytics v8.4.71 is mainly a release-tag update 📦, but the most important underlying change is a major refactor of the official C++ examples that makes YOLO deployment in C++ much more unified, easier to use, and more capable across backends 🚀
Release bump to 8.4.71 🔖
8.4.70 to 8.4.71.Big C++ examples refactor landed in this release 🛠️
examples/cpp/ structure instead of being scattered across many separate folders.Much broader C++ task support 🎯
Automatic task and model behavior detection 🤖
More consistent C++ developer experience 🧰
yolo_<backend>--model and --sourceUltralytics Platform GPU docs expanded ☁️
For most users: this release is mostly about packaging and delivery** ✅
v8.4.71 helps tools and environments track the newest Ultralytics build correctly.For C++ developers: this is the real headline** 💡
For teams using multiple model types: better flexibility** 🔄
For YOLO26 adoption: stronger deployment support** 🚀
For Ultralytics Platform users: more cloud hardware choices** ⚡
In short: the tag PR itself is just a version bump, but v8.4.71 packages a meaningful C++ usability upgrade and refreshed platform GPU documentation 🎉
Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.4.70...v8.4.71
v8.4.70 is a reliability-focused release 🛠️ led by an important fix for multi-GPU training, plus safer checkpoint loading, a plotting fix for segmenta
v8.4.70 is a reliability-focused release 🛠️ led by an important fix for multi-GPU training, plus safer checkpoint loading, a plotting fix for segmentation results, and clearer docs around YOLO26 heatmaps, training outputs, and deployment 📚
🐛 Fixed DDP AMP synchronization in multi-GPU training in PR #24810 by @LZ-QWQ
✅ Safer checkpoint loading with clearer errors in PR #24857 by @glenn-jocher
state_dict files or some older YOLOv5-style module references.AttributeError or KeyError.🎨 Fixed Results.plot() for masks-only outputs in PR #24635 by @ahmet-f-gumustas
🔥 Heatmaps guide was rewritten around real tracking use cases in PR #24815 by @raimbekovm
📁 Documented the save_dir training argument in PR #24831 by @raimbekovm
🧩 Small quality-of-life fixes in docs and config
region argument from the instance segmentation guide in PR #24833 by @raimbekovmfigsize typing to floats in Solutions config in PR #24832 by @raimbekovm🚀 More reliable distributed training
🛡️ Fewer cryptic failures
🎯 Better support for segmentation workflows
📚 Clearer guidance for real-world YOLO26 usage
⚙️ Smoother day-to-day experience
In short: v8.4.70 is mainly a stability and usability release ✅ with the standout improvement being a critical fix for consistent AMP behavior in distributed multi-GPU training.
models.yolo by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/24857Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.4.69...v8.4.70
Ultralytics v8.4.69 is a small but important reliability update 🔒 that improves safe checkpoint loading, especially for some YOLO11 and YOLOv8 segment
Ultralytics v8.4.69 is a small but important reliability update 🔒 that improves safe checkpoint loading, especially for some YOLO11 and YOLOv8 segmentation, pose, OBB, and classification models.
🔐 Safe loading now supports certain getattr references inside checkpoints
getattr, which were previously blocked when safe loading was enabled.🤖 Fixes safe loading for several model types
🧩 Added restricted support for enum-based references
🛡️ Still keeps the load process restricted
getattr use, the change adds a limited internal handler that only permits approved cases.📦 Version bump
8.4.68 to 8.4.69.✅ Fewer loading/export errors with safe mode enabled
ULTRALYTICS_SAFE_LOAD=1 should now be able to load and export affected checkpoints more smoothly.🔒 Better security without losing usability
🚀 Improves deployment and export workflows
👥 Good quality-of-life fix for everyday users
📌 Low-risk, high-value release
If helpful, I can also provide a one-paragraph changelog version or a more technical developer-focused summary.
getattr by @Y-T-G in https://github.com/ultralytics/ultralytics/pull/24852Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.4.68...v8.4.69
Security fix for similarity search UI 🛡️
Ultralytics v8.4.68 is mainly a training-quality and reliability release, led by a major improvement for semantic segmentation: you can now use class weighting to better handle imbalanced datasets 🎯🧠
Semantic segmentation now supports class weights via cls_pw 🖼️⚖️
Default semantic training behavior stays the same 🔒
cls_pw is not set, or is 0, Ultralytics keeps the previous behavior.Better weighting logic for segmentation 🧮➡️🧠
More robust handling of missing or unreadable labels/masks 🛠️
Detection classification loss can now apply class weights too 📦
Several production stability fixes 🚑
Safer model loading on older PyTorch versions 🔐
torch<2.5 instead of erroring out.Security fix for similarity search UI 🛡️
Docs and platform updates 📚
Better semantic segmentation on imbalanced datasets 🌍
Safer upgrade path ✅
cls_pw, most users can upgrade without changing existing results.More control for advanced users 🎛️
Improved robustness in production 🏭
Broader hardware confidence 💻
Better security and compatibility 🔒
Overall, v8.4.68 is a practical quality release with its biggest win being more capable and customizable semantic segmentation training, plus several important fixes that make Ultralytics more stable and dependable in real-world use 🚀
Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.4.67...v8.4.68
v8.4.67 is a small but meaningful release focused on safer model loading 🔒, plus an important TT100K dataset label fix 🛣️ and a batch of documentation
v8.4.67 is a small but meaningful release focused on safer model loading 🔒, plus an important TT100K dataset label fix 🛣️ and a batch of documentation/link cleanups 📚.
🔒 New opt-in safe model loading via ULTRALYTICS_SAFE_LOAD (@glenn-jocher, PR #24829)
ULTRALYTICS_SAFE_LOAD=true, that enables a safer way to load model checkpoints.weights_only=True behavior and only rebuilds known Ultralytics and PyTorch model classes from an automatically generated allow-list.SafeUnpickler approach.🛣️ TT100K dataset class list corrected to the official 221-category set (@glenn-jocher, PR #24718)
TT100K.yaml dataset config had duplicated and incorrect class names.📘 Rust inference docs updated to ultralytics-inference 0.0.21 (@onuralpszr, PR #24825)
🔗 Many documentation links were refreshed and corrected (@glenn-jocher, PR #24824 and #24817)
🔐 Safer checkpoint loading for security-conscious users
♻️ Better future compatibility with PyTorch
weights_only loading by default.🚫 Less chance of silent dataset errors
🧰 Minimal disruption for most users
📚 Clearer onboarding and fewer broken docs
If you want, I can also provide a one-paragraph release note version or a developer-focused summary of v8.4.67.
Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.4.66...v8.4.67
Ultralytics v8.4.66 improves built-in NVIDIA GPU monitoring reliability 🔋, adds better semantic segmentation class filtering and metrics 🧠, and enhanc
Ultralytics v8.4.66 improves built-in NVIDIA GPU monitoring reliability 🔋, adds better semantic segmentation class filtering and metrics 🧠, and enhances CoreML semantic exports on iPhone with full-resolution outputs 📱.
Main update: NVIDIA GPU monitoring is now more reliable 🚀
nvidia-ml-py was added as a standard dependency by @glenn-jocher.Semantic segmentation got a meaningful feature upgrade 🎯
classes filter now works properly for semantic segmentation during:
single_cls=True in semantic segmentation.Semantic segmentation metrics are more accurate 📈
CoreML semantic exports now keep full-resolution maps 📱✨
Updated iPhone CoreML benchmark docs 📊
TensorRT 11 export flow was simplified ⚙️
half argument was removed from ONNX precision conversion.Documentation improvements across YOLO26 guides 📝
CI and packaging cleanup 🛠️
nvidia-ml-py is now a standard dependency, extra manual installs were removed from CI and Docker setup.More dependable GPU stats for users with NVIDIA hardware ✅
Users should see fewer issues with built-in GPU monitoring, logging, and hardware-aware utilities because the required NVIDIA package is now installed up front instead of handled on the fly.
Better usability for semantic segmentation workflows 🎨
If you work with semantic segmentation and only care about certain classes, filtering now behaves much more intuitively during prediction and evaluation.
More trustworthy segmentation metrics 📏
Reported semantic segmentation scores are now less misleading, especially on datasets where some classes are absent or filtered out.
Sharper CoreML semantic results on iPhone and Apple devices 🍎
Semantic masks should look much cleaner and less blocky, improving visual quality for mobile deployment.
Cleaner export experience for TensorRT 11 ⚡
Developers exporting models for NVIDIA deployment get a simpler and more maintainable precision conversion workflow.
Clearer docs for a broader audience 📚
New and experienced users alike should find YOLO26 testing, training, and deployment guidance easier to follow.
Overall, v8.4.66 is a practical quality-focused release: it strengthens GPU monitoring, improves semantic segmentation behavior, and brings better mobile export quality without changing the core user workflow.
classes filter to semantic segmentation task by @lmycross in https://github.com/ultralytics/ultralytics/pull/24806nvidia-ml-py to pyproject.toml by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/23922Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.4.65...v8.4.66
v8.4.65 is mainly a mobile inference performance release 📱⚡, with the biggest upgrade improving QNN exports for Snapdragon NPUs and speeding up semant
v8.4.65 is mainly a mobile inference performance release 📱⚡, with the biggest upgrade improving QNN exports for Snapdragon NPUs and speeding up semantic segmentation outputs by moving costly work inside the model graph.
🚀 Major QNN export performance upgrade in PR #24790 by @glenn-jocher
QNNModel instead of fragile post-export graph editing.🧠 Semantic segmentation exports are much faster on-device
ClassMapModel wrapper handles this behavior during export.📱 Better mobile runtime compatibility
📉 Measured impact highlighted in docs
🛠️ RKNN export improvements
🐳 Docker build reliability improvements
📚 Docs and export guidance improvements
ultralytics-inference version 0.0.19.✅ CI and workflow fixes
main.⚡ Faster mobile inference on Snapdragon devices
Users deploying YOLO on Qualcomm hardware should see lower overhead and better real-world efficiency, especially in camera-based apps where image buffers are already channel-last.
🧩 Less app-side postprocessing work
By returning semantic class maps directly from exported models, apps no longer need to spend as much CPU time decoding huge segmentation outputs.
📈 Better and more stable semantic segmentation performance
This is especially important for real-time mobile experiences, where unpredictable postprocessing delays can cause lag or jitter.
🔒 More robust export pipeline
Replacing manual ONNX graph surgery with clean wrapper modules makes exports easier to maintain and less error-prone over time.
📱🍎 Benefits extend beyond QNN CoreML semantic exports also gain from the same in-graph class-map idea, so Apple-device deployment gets a speed boost too.
🛠️ Improved reliability for deployment workflows Better RKNN testing, cleaner Docker builds, and more dependable docs publishing all reduce friction for developers working across edge and production environments.
👥 Broad user impact
In short: v8.4.65 is a strong deployment-focused release 🎉, with the standout improvement being faster, more hardware-friendly QNN exports and smarter semantic segmentation outputs for mobile AI.
Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.4.64...v8.4.65
Ultralytics v8.4.64 makes tracking easier and more reliable 🎯—especially with new auto-downloaded YOLO26 ReID ONNX encoders—while also improving QNN e
Ultralytics v8.4.64 makes tracking easier and more reliable 🎯—especially with new auto-downloaded YOLO26 ReID ONNX encoders—while also improving QNN export, multi-GPU training stability, logging, and export robustness 🚀
Tracking got a major usability upgrade with auto-downloaded YOLO26 ReID ONNX encoders 🤖📦
You can now use built-in tracking ReID models like yolo26n-reid.onnx through yolo26x-reid.onnx without manually downloading or exporting them first.
Tracking ReID matching is more reliable when appearance features are missing 🔍
TrackTrack now treats missing embeddings as “unknown” instead of assuming they are bad matches, so it can fall back to motion cues more intelligently.
QNN export for Qualcomm devices was significantly improved 📱⚡
QNN models now export as a single self-contained *_qnn.onnx file instead of a folder, with metadata embedded inside. The release also fixes architecture mapping, improves backend loading compatibility, and updates quantization behavior for better Snapdragon deployment.
YOLO26x distributed training stability was fixed 🧠🖥️
A DDP deadlock issue affecting some multi-GPU training runs was resolved by restoring handling for unused parameters in conditional branches.
MLflow failures no longer crash training 📉🛡️
If MLflow tracking setup or logging fails, training now continues instead of aborting the run.
TensorFlow export subprocess calls are safer 🔐
Edge TPU and TensorFlow.js export commands now avoid shell-based path handling issues, reducing problems with unusual file paths and improving security.
Progress/logging output was cleaned up 🖥️✨
Fixed premature 100% progress display and console log duplication issues, especially useful in platform or remote logging environments.
Version checking is more accurate ✅
parse_version() now consistently returns 3-part version tuples, fixing incorrect version comparisons like 6.0 vs 6.0.0.
FP16 quantization/export reliability was improved ⚙️
ONNX mixed-precision conversion now uses the correct input name and sample input shape, helping TensorRT/ModelOpt workflows work more consistently.
Docs were refreshed across several areas 📘
Updates include clearer tracking/ReID docs, QNN docs, Conda install guidance, hyperparameter tuning explanations, TrackZone behavior, custom trainer checkpoint loading, OpenVINO benchmark references, and terminal visualization guidance.
Easier multi-object tracking setup 🚀
The headline change removes a common setup headache: users can now enable tracking ReID with ready-made YOLO26 ONNX encoders directly, making advanced tracking more accessible to both beginners and production teams.
Better tracking quality in difficult scenes 🎥
The ReID fallback improvement should reduce bad associations when objects are briefly occluded or appearance features are unavailable.
Simpler Qualcomm deployment 📱
QNN export is now easier to manage and deploy thanks to the single-file output format and improved compatibility across Snapdragon targets.
More dependable training at scale 🖥️
Teams training larger YOLO26 models on multiple GPUs should see fewer hangs and more stable runs.
Fewer pipeline interruptions 🛠️
MLflow and export-related fixes help ensure optional integrations do not stop core training or deployment workflows.
Cleaner user experience ✨
Console logging and progress bar fixes make training output easier to trust and monitor, especially in shared platforms and dashboards.
Lower friction for installation and documentation 📚
Updated guides make setup, tuning, export, and custom workflows easier to understand and reproduce.
In short, v8.4.64 is a tracking-focused quality-of-life release with especially strong benefits for users working with YOLO26 tracking, Qualcomm QNN deployment, and stable training/export pipelines 🚀
parse_version to always return a 3-tuple by @bujna94 in https://github.com/ultralytics/ultralytics/pull/24680openvino==2026.2.0 by @lakshanthad in https://github.com/ultralytics/ultralytics/pull/24691Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.4.63...v8.4.64
FP16 and INT8 exports are still available even with TensorRT 11’s breaking changes.
🚀 Ultralytics v8.4.63 mainly adds TensorRT 11 export support with FP16 and INT8 quantization via NVIDIA ModelOpt, while also expanding built-in multi-object tracking with several new tracker options and improving reliability, performance, and docs.
🧠 Major export upgrade: TensorRT 11 support
fp32fp16int8⚡ Better quantization workflow for modern NVIDIA deployment
🎯 Big tracking expansion: 4 new built-in trackers
ocsort.yamldeepocsort.yamlfasttrack.yamltracktrack.yaml🏃 Tracking docs and selection guidance improved
📹 Video stream loading is safer
⚡ AI Gym pose workflow is faster
🧪 Validation mixed precision handling simplified
📝 Documentation improvements
🚀 TensorRT 11 users can export again
⚙️ Future-proofs NVIDIA deployment
💾 Smaller, faster engines remain accessible
📦 Dynamic INT8 export support is especially useful
👀 Tracking becomes more flexible for real-world scenarios
🔒 More stable long-running applications
⚡ Small but meaningful performance wins
Overall, v8.4.63 is a strong deployment-focused release 📦—with the headline improvement being restored and modernized TensorRT 11 export support, plus a major boost to tracking capabilities and several reliability/performance refinements.
isolated_model and isolated_task_model into isolated_model_path by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/24742Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.4.62...v8.4.63
🛡️ v8.4.62 is mainly a reliability release focused on preventing trained models from being lost at the end of training, with additional improvements t
🛡️ v8.4.62 is mainly a reliability release focused on preventing trained models from being lost at the end of training, with additional improvements to Platform docs, dataset/API documentation, test stability, and CI efficiency.
🚨 Major training fix: checkpoints are no longer discarded just because EMA hits NaN/Inf during save checks
✅ Validation is now safer during AMP training
🧪 New test coverage for fp16 overflow checkpoint handling
📘 Big Ultralytics Platform docs refresh
🔗 Fixed broken COCO evaluation links
🧪 Less flaky data-related tests
⚡ Lean CI improvements
💾 Prevents losing trained models
🔒 Improves training stability and trustworthiness
🚀 Better experience for common training setups
📚 More accurate docs for the Ultralytics Platform
🧰 Improved developer and CI reliability
🌍 Cleaner external documentation links
Overall, v8.4.62 is not a major model-feature release, but it is a high-value stability update 🛠️—especially for anyone training YOLO models with mixed precision and expecting reliable checkpoint saves.
Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.4.61...v8.4.62
v8.4.61 is mainly a stability and export reliability release 🛠️, led by an important fix for INT8 ONNX export failures and another fix for read-only `
v8.4.61 is mainly a stability and export reliability release 🛠️, led by an important fix for INT8 ONNX export failures and another fix for read-only onnx2tf patching, with additional improvements to CI, export testing, docs accuracy, and platform documentation.
🚨 Fixed INT8 ONNX export crashes on small calibration datasets in PR #24721 by @glenn-jocher
🔒 Fixed read-only onnx2tf patching issues in PR #24721 by @glenn-jocher
🧪 Stronger export validation in CI
yolo26n.pt, helping catch export problems earlier.🤖 TensorRT compatibility improvement
end2end export when using older TensorRT versions that do not support it.🎯 SAM duplicate-mask cleanup fix
🧠 Semantic segmentation support made clearer across the product
task=semantic as a valid option.📚 Large documentation and accuracy refresh
✅ More reliable model export workflows
🚀 Fewer production export failures
onnx2tf and ONNX calibration fixes target bugs already seen in real-world error tracking, so this release should reduce export breakages in actual deployments.🧪 Better confidence in deployment formats
📦 Improved compatibility on specialized hardware
🧹 Cleaner and more accurate user experience
🌍 Better guidance for a broad user base
In short, v8.4.61 is less about new models and more about making YOLO26 deployment safer, smoother, and more production-ready 🔧📈
tip case in docs by @lakshanthad in https://github.com/ultralytics/ultralytics/pull/24669Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.4.60...v8.4.61
Ultralytics v8.4.60 is mainly about adding ONNX INT8 export 🎉, making it easier to create smaller, faster deployment models with built-in calibration
Ultralytics v8.4.60 is mainly about adding ONNX INT8 export 🎉, making it easier to create smaller, faster deployment models with built-in calibration support, while also including a few helpful export, training, and documentation fixes.
🚀 Major new feature: ONNX int8=True export
data for calibration dataset selection and fraction for using only part of the dataset.*_int8.onnx.🔄 Shared INT8 calibration pipeline
📘 Much better ONNX export documentation
⚙️ RKNN export now supports the standard half argument
half=True, and this becomes the default floating-point path for supported Rockchip hardware.🐛 Segmentation training fix for polygons on image borders
segment2box ensures polygon points lying exactly on image edges are no longer dropped.📝 Auto-annotate docs updated
output_dir for auto-annotation was corrected.🧹 Docs metadata cleanup
🎯 Faster and lighter ONNX deployment
🛠️ Simpler export workflow
🔒 More reliable maintenance and consistency
📈 Better user experience for deployment
🤖 Improved hardware export support
half=True update helps Rockchip deployments behave more predictably and aligns them better with common export expectations.🖼️ More accurate segmentation training
Overall, v8.4.60 is a deployment-focused release 🌍, with ONNX INT8 export as the standout improvement and several supporting fixes that improve reliability, documentation, and hardware export consistency.
ultralytics 8.4.60 ONNX INT8 export by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/24666Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.4.59...v8.4.60
Ultralytics v8.4.59 improves Rockchip RKNN export with safer and clearer INT8 quantization support, making deployment on Rockchip edge devices more re
Ultralytics v8.4.59 improves Rockchip RKNN export with safer and clearer INT8 quantization support, making deployment on Rockchip edge devices more reliable and easier to use 🚀
✅ RKNN INT8 calibration file is now validated before export proceeds
🧠 Simpler RKNN INT8 workflow for users
data=... during export.dataset.txt calibration file automatically.dataset.txt stays an internal implementation detail, not something users are expected to manage directly.⚡ New RKNN INT8 export support introduced in this release cycle
int8=Truedata=...fraction=...📱 Expanded support for INT8-only Rockchip chips
int8=True.📚 Documentation and export tables updated
int8=True is requireddata and fraction for calibration🛡️ Better RKNN API error handling
More reliable RKNN exports 🔒
Easier INT8 quantization for everyone 🙌
data=coco8.yaml or another dataset YAML, without learning RKNN-specific file conventions.Better support for edge deployment on Rockchip hardware 📦
Cleaner user experience 🧹
Potential performance and compatibility gains 🚀
Lower risk of misconfiguration ⚠️
If helpful, I can also provide a 1-paragraph release note version or a developer-focused changelog summary for v8.4.59.
ultralytics 8.4.59 Rockchip RKNN INT8 export option by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/24656Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.4.58...v8.4.59
Ultralytics v8.4.58 is mainly a reliability-focused release 🛠️ that improves how export formats are tested in CI, making model export support more sta
Ultralytics v8.4.58 is mainly a reliability-focused release 🛠️ that improves how export formats are tested in CI, making model export support more stable, isolated, and easier to maintain—especially for specialized formats.
🚦 Export testing was redesigned around isolated environments
🧩 Each export format now declares its own environment
export_formats() registry now includes environment metadata for every export target.🧪 New --export-env test selection
⚙️ New automated isolated export job
IsolatedExports CI workflow now builds dedicated virtual environments and runs smoke tests plus export tests inside them.📦 Export dependencies were split into smaller groups
export-baseexport-tensorflowexport-coremlexport-executorchexport-deepxexport-legacy-torch✅ Added checks to prevent export registry mistakes
🤖 RKNN export test added
🐳 Docker cleanup improved
🔒 More reliable exports
🚀 Faster and leaner development workflows
🧹 Simpler maintenance for the Ultralytics team
🌍 Better support for specialized deployment targets
📌 No major new model architecture in this release
🧠 Good news for production users
If you'd like, I can also turn this into a shorter changelog-style summary or a more technical developer-focused breakdown.
ultralytics 8.4.58 Isolated CI export environments by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/24649Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.4.57...v8.4.58
🚀 Ultralytics v8.4.57 is mainly a Python 3.13 readiness release, making CI and export workflows more robust across platforms, while also adding a mean
🚀 Ultralytics v8.4.57 is mainly a Python 3.13 readiness release, making CI and export workflows more robust across platforms, while also adding a meaningful RT-DETR CoreML export upgrade and a semantic segmentation mask alignment fix.
🐍 Python 3.13 CI is now the main focus of this release
🛡️ Export and slow-test CI became much more resilient
🚫 Python 3.13 compatibility guards were added for unsupported export stacks
📦 RT-DETR CoreML export improved significantly
🎯 Semantic segmentation masks are now better aligned
📚 Docs and platform updates
✅ Better future-proofing for users on new Python versions
🔍 Much easier debugging for maintainers and advanced users
📤 Safer export experience across many backends
🍎 RT-DETR deployment on Apple ecosystems gets more practical
🖼️ Semantic segmentation results should look cleaner
🌐 Easier migration into the Ultralytics Platform
Overall, v8.4.57 is less about introducing a brand-new model and more about stability, compatibility, and deployment polish 🛠️✨—especially for Python 3.13, export workflows, and RT-DETR/CoreML users.
test_export by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/24609ultralytics 8.4.57 Python 3.13 CI by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/24640Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.4.56...v8.4.57
Ultralytics v8.4.56 improves QNN export reliability by fixing compatibility with newer onnxruntime-qnn packages, especially on Linux x86-64. 🚀
Ultralytics v8.4.56 improves QNN export reliability by fixing compatibility with newer onnxruntime-qnn packages, especially on Linux x86-64. 🚀
Fixed QNN export for built-in provider wheels 🔧
Ultralytics now detects whether onnxruntime-qnn already includes QNNExecutionProvider internally, instead of always trying to register it as a separate plugin.
Avoids export failures on some Linux setups 🐧
This prevents a known failure on Linux x86-64 wheels where plugin registration could crash with an undefined symbol error.
Supports both QNN packaging styles 📦
The export flow now works with:
Smarter QNN session creation 🧠
If the provider is built in, Ultralytics now launches the ONNX Runtime session directly with the QNN provider. If it is plugin-based, it keeps the previous registration path.
Documentation/comments updated for clarity 📝
Internal descriptions were updated to reflect that QNN may be exposed either as a plugin or as a built-in ONNX Runtime provider.
Makes QNN export more robust ✅
Users exporting models to Qualcomm QNN should see fewer environment-specific errors and smoother exports.
Improves compatibility with newer runtime wheels 🔄
This is especially helpful as onnxruntime-qnn packaging evolves across platforms and releases.
Reduces setup friction for deployment 📱
If you are exporting YOLO models like YOLO26 for Qualcomm and edge hardware workflows, this change lowers the chance of export-time breakage.
No major model architecture changes in this release 🤖
v8.4.56 is mainly a stability and export compatibility update, not a new model or training feature release.
Most relevant for advanced deployment users ⚡
If you use standard training and inference, this release may not change much day to day. But for users targeting QNN / Qualcomm deployment pipelines, it should have a meaningful positive impact.
This release is a small but important deployment fix: it makes QNN export smarter and more portable by handling both old and new onnxruntime-qnn layouts correctly, with the biggest benefit for Linux-based export environments.
Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.4.55...v8.4.56
Ultralytics v8.4.55 is a reliability-focused release that mainly fixes ONNX/QNN export environments so existing onnxruntime variants are no longer acc
Ultralytics v8.4.55 is a reliability-focused release that mainly fixes ONNX/QNN export environments so existing onnxruntime variants are no longer accidentally overwritten, while also improving deployment docs, especially for Hailo and YOLO26 task support 🚀
🛠️ Major export fix for ONNX/QNN environments
onnxruntime package variants such as:
onnxruntime-qnnonnxruntime-directmlonnxruntime-openvinoonnxruntime-trainingonnxruntime, which could overwrite specialized builds and break export or inference.📦 Safer QNN export workflows
📘 New Hailo deployment guide
model.export() target, but can still be used through the external Hailo toolchain.🧠 Broader and clearer YOLO26 task documentation
⚡ Faster and more stable CI testing
pytest-xdist.📈 Improved Comet documentation
COMET_* environment variable docs and clarified online/offline logging behavior.🔄 Minor maintenance updates
actions/cache@v4 to @v5.✅ More reliable exports for advanced runtimes
📱 Better support for edge and hardware-specific deployment
🧩 Less confusion around segmentation tasks
🛡️ Better overall stability
🌍 Easier adoption of YOLO26 capabilities
In short: v8.4.55 is mainly a stability and compatibility release 🔧—especially valuable for users exporting to ONNX/QNN or working with specialized runtime builds—while also making deployment and YOLO26 task support clearer for the broader community.
ultralytics 8.4.55 Accept onnxruntime variants in ONNX export requirements check by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/24611Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.4.54...v8.4.55
Ultralytics v8.4.54 is led by a big new deployment feature: local Qualcomm QNN export for Snapdragon devices 📱⚡, plus several quality-of-life improvem
Ultralytics v8.4.54 is led by a big new deployment feature: local Qualcomm QNN export for Snapdragon devices 📱⚡, plus several quality-of-life improvements for Solutions, export docs, tuning compatibility, and data-loading reliability.
🚀 New qnn export format for Qualcomm Snapdragon devices
format="qnn".from ultralytics import YOLO
model = YOLO("yolo26n.pt")
model.export(format="qnn")
📦 New QNN backend support for inference/validation workflows
📘 Major export documentation cleanup and standardization
🖼️ imgsz added to all Ultralytics Solutions
imgsz argument, giving users direct control over inference input size.🛠️ Better robustness for segmentation and image caching
(H, W, 1) instead of plain single-channel arrays..npy image cache issues that could cause grayscale models to accidentally load RGB data.🧠 Tracking utility optimization
🎯 Ray Tune compatibility improvement
🧾 Small but useful documentation updates
show_labels and show_conf..jpeg2000 as a supported image format.🔬 Semantic segmentation loss dtype fix
📱 Big win for edge/mobile deployment
⚡ Faster path to production on Snapdragon
🧩 More consistent deployment experience
🎛️ More control in packaged Solutions
imgsz to Solutions gives users a simple way to trade off speed vs. accuracy, just like in normal YOLO inference.🛡️ More reliable training and inference
🔄 Better compatibility with modern tooling
👀 Cleaner debugging and validation visuals
show_labels and show_conf options make validation outputs easier to customize for analysis.Overall, v8.4.54 is a strong release for anyone interested in device deployment, especially on Qualcomm Snapdragon, while also tightening up reliability and usability across the broader Ultralytics ecosystem 🚀
embedding_distance with NumPy by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/24572_gaussian_smooth changes by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/24575imgsz support to all Ultralytics Solutions by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/24579cv2.imread by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/24582ultralytics 8.4.54 Qualcomm QNN export by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/24591Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.4.53...v8.4.54
Ultralytics v8.4.53 mainly improves training reliability on NVIDIA GPUs by automatically recovering from more CUDA memory-related failures, while also
Ultralytics v8.4.53 mainly improves training reliability on NVIDIA GPUs by automatically recovering from more CUDA memory-related failures, while also polishing semantic segmentation stability, documentation clarity, and CI robustness 🚀
Smarter GPU memory recovery during training 🧠⚡
The biggest change in this release, from PR #24569 by @glenn-jocher, expands the existing auto-retry logic for large-batch GPU training.
CUDNN_STATUS_INTERNAL_ERRORSemantic segmentation validation is more reliable 🖼️✅
PR #24552 fixes issues in semantic validation so metrics are preserved correctly after evaluation.
NaN-style metric problems in sparse-class casesSegmentation loss is numerically more stable 🛠️📈
PR #24554 improves Dice loss stability by forcing key intermediate calculations to use float32.
Much better docs for prediction outputs, especially semantic segmentation 📚
PR #24558 is a major documentation improvement.
Results object for each tasksemantic_mask output in detailBroader YOLO26 and semantic task visibility across docs and metadata 🌍
Several updates improve how Ultralytics presents current capabilities:
CI and test workflow became more stable 🧪
Multiple changes reduce avoidable test failures:
Small but useful documentation and packaging fixes ✨
Fewer frustrating training crashes on GPUs 💪
If you train with aggressive batch sizes, this release can help Ultralytics recover automatically from more real-world CUDA memory failures instead of stopping unexpectedly.
Better experience for users pushing hardware limits 🚀
Large-batch training on single GPUs should now be more forgiving, especially when failures come from CUDA backend libraries rather than a plain out-of-memory exception.
More dependable semantic segmentation workflows 🧠
Users working with semantic segmentation should see:
Easier to understand model outputs 🔍
The new docs make it much clearer what prediction results contain for each task, which is especially helpful for:
ResultsBetter confidence in production use ✅
The “Production/Stable” metadata and improved CI signal a continued push toward reliability for real deployments.
Low risk release with practical quality-of-life improvements 🙂
This is not a major architecture or model release, but it delivers meaningful gains in stability, clarity, and usability, especially for training and semantic segmentation users.
nt_per_image printing issue by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/24552track_high_thresh in track docs by @raimbekovm in https://github.com/ultralytics/ultralytics/pull/24559ultralytics 8.4.53 Retry CUDA backend memory errors during training by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/24569Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.4.52...v8.4.53
🚀 Ultralytics 8.4.52 is a major release centered on adding full YOLO26 semantic segmentation support, plus several reliability, export, visualization,
🚀 Ultralytics 8.4.52 is a major release centered on adding full YOLO26 semantic segmentation support, plus several reliability, export, visualization, and documentation improvements.
🆕 New semantic task added across Ultralytics
🧠 New YOLO26 semantic segmentation models
-sem suffix, such as:
yolo26n-sem.ptyolo26s-sem.ptyolo26m-sem.ptyolo26l-sem.ptyolo26x-sem.pt🏗️ Complete semantic segmentation pipeline introduced
SemanticSegmentationModelSemanticSegmentationTrainerSemanticSegmentationValidatorSemanticSegmentationPredictor🖼️ Semantic masks now appear directly in inference results
SemanticMask result type.📦 Built-in semantic segmentation datasets added
🔁 Augmentation pipeline upgraded for semantic masks
📤 Export and deployment support extended
nms=True is automatically disabled for semantic models, since it does not apply to this task.🧪 Much broader testing coverage
🛠️ Other notable fixes in this release
0.0 GFLOPs. 📏show_labels and show_conf, making crowded scenes easier to inspect. 👀🎯 Biggest impact: YOLO26 can now do semantic segmentation natively
⚡ Easier adoption for existing YOLO users
🧩 Broader dataset compatibility
📉 More reliable metrics and logging
🔍 Better debugging and visualization
🚀 Stronger deployment readiness
📚 Improved docs and setup guidance
In short, v8.4.52 is a feature-heavy release led by the launch of YOLO26 semantic segmentation 🎉. For many users, this is the headline change and a meaningful expansion of what Ultralytics can now handle out of the box.
get_flops dtype mismatch on half-precision models by @ahmet-f-gumustas in https://github.com/ultralytics/ultralytics/pull/24504on_fit_epoch_end in final_eval overwriting last epoch metrics by @deependujha in https://github.com/ultralytics/ultralytics/pull/24530zensical 0.0.43 by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/24533ultralytics 8.4.52 YOLO Semantic Segmentation models by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/24518Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.4.51...v8.4.52
Ultralytics v8.4.51 focuses mainly on better training traceability and clearer deployment/docs updates 📦📝, with the most important change adding the G
Ultralytics v8.4.51 focuses mainly on better training traceability and clearer deployment/docs updates 📦📝, with the most important change adding the Git commit message to training metadata so models are easier to track, reproduce, and audit.
Training metadata now includes the Git commit message 🧾
The headline update from @glenn-jocher adds the current commit subject into:
git.messagegitCommitMessageGitRepo.messageMore robust Git metadata handling 🔧
Git repository parsing was improved to better read metadata from Git internals, including shared/worktree-style layouts. This helps Ultralytics capture version information more reliably during training.
Major augmentation pipeline refactor 🛠️
A substantial internal refactor by @Laughing-q introduced a more unified transform system:
BaseTransform now standardizes how image, instance, and semantic-mask transforms are appliedMosaic, MixUp, CutMix, CopyPaste, RandomPerspective, RandomFlip, and LetterBox were reorganized around this shared structureOpenVINO docs updated with YOLO26 benchmarks 🚀
The OpenVINO documentation now highlights YOLO26 benchmark results instead of older YOLO11 benchmarks, with refreshed performance data across newer Intel CPUs, GPUs, and NPUs.
DeepX export documentation expanded 📤
DeepX was added to the export formats table, with supported export arguments and output folder behavior documented more clearly.
RT-DETR inference tuning guidance added ⚡
Docs now explain that users can reduce query count for faster RT-DETR inference, helping users trade a bit of accuracy for lower latency when needed.
YOLOE export behavior clarified ⚠️
The docs now clearly warn that exported YOLOE models are static: once exported, prompt-based class configuration is baked into the model and cannot be changed later.
Ultralytics Platform GPU docs refreshed ☁️
Platform docs now reflect:
Test/CI compatibility improvement for Axelera export 🧪
Axelera export tests are now limited to supported PyTorch versions, reducing false failures in CI.
General documentation cleanup 📚
Smaller updates include a fixed DeepX link, README simplification, removal of old Weglot docs overrides, and wording/casing polish across docs.
Easier experiment tracking and reproducibility 🔍
Adding the Git commit message makes it much easier to tell what exact code change produced a trained model, especially when many experiments are run close together.
Better debugging and collaboration 🤝
Teams using local training or the Ultralytics Platform can now connect checkpoints and cloud runs to a human-readable commit description, not just a hash.
Stronger foundations for future augmentation work 🧱
The transform refactor is mostly an internal improvement, but it should make augmentations more consistent, easier to maintain, and safer to extend in future releases.
Clearer deployment decisions for users 📈
Updated YOLO26 OpenVINO benchmarks, DeepX export docs, and RT-DETR speed tips help users choose faster deployment settings with more confidence.
Fewer surprises in export workflows ✅
The YOLOE warning helps users avoid exporting a model and later discovering that prompt changes no longer work.
Improved platform transparency 💳
Updated GPU availability and pricing docs help users better plan cloud training costs and choose the right hardware tier.
Overall, v8.4.51 is less about new end-user model features and more about making training runs easier to understand, reproduce, and deploy reliably 🌟
2.8.0<=torch<2.12.0 in Tests CI by @lakshanthad in https://github.com/ultralytics/ultralytics/pull/24487BaseTransform and unify augmentation pipeline by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/24434ultralytics 8.4.51 Add Git commit message to training metadata by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/24505Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.4.50...v8.4.51
Ultralytics 8.4.50 is mainly a deployment-focused release 🚀, led by a new DeepX export and inference integration that makes it easier to run YOLO mode
Ultralytics 8.4.50 is mainly a deployment-focused release 🚀, led by a new DeepX export and inference integration that makes it easier to run YOLO models on DeepX NPU edge hardware, along with a few quality-of-life fixes for tuning reliability, mixed-precision model fusion, and RT-DETR documentation.
New DeepX export support added 🧠⚡
You can now export Ultralytics models directly with format="deepx" as part of the normal export workflow.
.dxnn format.config.json for preprocessing and calibrationmetadata.yaml for model informationint8=True.coco128.yaml.New DeepX inference backend added 🔌
Exported DeepX models can now be loaded back into Ultralytics for inference through a dedicated backend, making deployment more integrated and easier to use.
DeepX documentation and references added 📘
This release includes a full DeepX integration guide, reference pages, export docs, and navigation updates so users can more easily learn and adopt the new workflow.
Hyperparameter tuning now reports failures more honestly 🛠️
The tuner no longer claims all iterations completed successfully when training runs actually failed.
best_hyperparameters.yaml files are no longer written when no valid result existsFix for dtype issues during Conv/BN and Deconv/BN fusion 🎯
Fusing models in float16 or bfloat16 now works correctly, avoiding dtype mismatch errors during optimization.
RT-DETR docs improved with eval_idx guidance 📉➡️⚡
The docs now explain how to reduce RT-DETR decoder layers at inference time to trade a small amount of accuracy for lower latency, without retraining.
Biggest impact: easier edge deployment on DeepX hardware 🚀
This is the headline feature of 8.4.50. Users targeting low-power embedded AI systems can now export YOLO models directly into a format optimized for DeepX NPUs, making deployment much smoother.
Better out-of-the-box hardware acceleration workflows ⚙️
By automatically handling INT8 export, calibration setup, and preprocessing config generation, the new DeepX integration reduces manual work and lowers the barrier to specialized edge deployment.
Improved production readiness for embedded use cases 🤖
This helps teams building applications like smart cameras, robotics, industrial automation, and other edge AI systems where speed and power efficiency matter.
More trustworthy tuning results ✅
Users running hyperparameter search will now get clearer feedback when experiments fail, helping avoid false confidence and bad downstream decisions.
More stable model optimization in reduced precision 🧩
The fusion fix is especially useful for advanced users working with float16 or bfloat16, improving reliability in optimized inference pipelines.
Helpful RT-DETR speed tuning guidance for latency-sensitive users ⏱️
While this is docs-only, it gives users a practical way to make RT-DETR faster in deployment scenarios without changing training.
dx_engineOverall, 8.4.50 is a strong deployment release, with DeepX integration as the standout addition 🌟—especially valuable for anyone moving Ultralytics models onto specialized edge AI hardware.
eval_idx instruction by @artest08 in https://github.com/ultralytics/ultralytics/pull/24465fuse_conv_and_bn and fuse_deconv_and_bn by @raimbekovm in https://github.com/ultralytics/ultralytics/pull/24480ultralytics 8.4.50 New Export Integration: Deepx by @ambitious-octopus in https://github.com/ultralytics/ultralytics/pull/23553Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.4.49...v8.4.50
Ultralytics v8.4.49 is mainly a stability and usability release 🛠️—it helps training fail earlier with clearer messages, recovers from corrupted cache
Ultralytics v8.4.49 is mainly a stability and usability release 🛠️—it helps training fail earlier with clearer messages, recovers from corrupted cached weights automatically, and includes several quality-of-life fixes across training, metrics, benchmarking, tuning, tracking, and docs.
🚨 Fail-fast training improvements from PR #24478 by @glenn-jocher:
.pt weight file is damaged or incomplete, Ultralytics now deletes it and re-downloads it instead of failing repeatedly.freeze is set too high and no trainable parameters remain, training now stops immediately with a direct, understandable error.FileNotFoundError naming the empty split.🧠 Better distributed training behavior with torch.compile from PR #24413 by @Y-T-G:
📈 Per-image F1 metric fix from PR #24444 by @fcakyon:
🧪 Hyperparameter tuning randomness fix from PR #24451 by @raimbekovm:
🛤️ Dataset relative path handling fix from PR #24429 by @auduntorp:
../ are now handled correctly in dataset file lists.⚙️ Benchmarking fix from PR #24450 by @Y-T-G:
data argument from being incorrectly forwarded into export benchmarking flows.🎯 Tracker API cleanup from PR #24457 by @Laughing-q:
frame_rate argument from BYTETracker and BOTSORT, making tracker setup simpler.🚀 CI / runner updates from PR #24436 by @glenn-jocher:
📚 Docs and reference improvements:
torch.compile, or hyperparameter evolution should see fewer odd edge-case failures.Overall, v8.4.49 is less about new models and more about making Ultralytics training and evaluation more robust, more predictable, and easier to troubleshoot ✅
static_graph=True for compile and remove find_unused_parameters by @Y-T-G in https://github.com/ultralytics/ultralytics/pull/24413data argument in benchmark call by @Y-T-G in https://github.com/ultralytics/ultralytics/pull/24450frame_rate in BYTETracker/BOTSORT by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/24457ultralytics 8.4.49 Fail fast on three common training failure modes by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/24478Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.4.48...v8.4.49
Ultralytics v8.4.48 is a stability-focused release that mainly improves training reliability on Ultralytics Platform and makes failure cases much clea
Ultralytics v8.4.48 is a stability-focused release that mainly improves training reliability on Ultralytics Platform and makes failure cases much clearer, with supporting fixes for benchmarking and reporting. 🛠️✅
(Top priority) Platform training edge-case fixes from PR #24431 by @glenn-jocher:
best/last) is saved, instead of failing later with confusing behavior. 💾ultralytics.* module, with guidance to retrain/use current official models. 📦Benchmark/export robustness fix from PR #24418 by @lakshanthad:
data is now only passed to export formats that actually support it, reducing export benchmark failures across formats. 📊Platform metrics correctness fix from PR #24425 by @mykolaxboiko:
training_complete.bestEpoch now reports the real best epoch (from early stopping), not just the final epoch trained. 🎯Docs improvement for multi-GPU training from PR #24422 by @artest08:
Maintenance/docs housekeeping
pip updates switched from daily to monthly (PR #24411 by @glenn-jocher). 📅bestEpoch, improving experiment tracking and decision-making. 📈data arg to export formats by @lakshanthad in https://github.com/ultralytics/ultralytics/pull/24418bestEpoch reporting wrong value at training_complete by @mykolaxboiko in https://github.com/ultralytics/ultralytics/pull/24425ultralytics 8.4.48 Fix Platform training edge cases by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/24431Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.4.47...v8.4.48
Ultralytics v8.4.47 is a reliability-focused release that fixes a key CLI bug for heatmap colormaps 🎨, while also improving RT-DETR post-processing, r
Ultralytics v8.4.47 is a reliability-focused release that fixes a key CLI bug for heatmap colormaps 🎨, while also improving RT-DETR post-processing, remote checkpoint loading, Edge TPU export behavior, and several stability edge cases across loaders and I/O ✅.
🔥 Most important (current PR #24219 by @raimbekovm): CLI colormap parsing fixed for Solutions Heatmap
colormap=cv2.COLORMAP_INFERNO now work as documented.cv2.<UPPERCASE_CONSTANT> values to numeric constants, with security protections still in place 🛡️.RT-DETR got a meaningful post-processing upgrade (PR #24403 by @artest08)
[cx, cy, w, h, score, class].Edge TPU export is now more robust and explicit (PR #24383 by @lakshanthad)
int8=True, Ultralytics now warns and auto-enables it early.Remote checkpoint URI support improved (PR #24395 by @glenn-jocher)
pretrained=ul://... now resolves like other remote model references 🌐.Multiple bug fixes across data loading, hub session, and solutions (PR #24397 by @glenn-jocher)
Safer image reading behavior (PR #24406 by @Y-T-G)
imread() now returns None on missing/unreadable files (matching common cv2.imread expectations), instead of crashing.YOLOE validation guardrail added (PR #23982 by @ahmet-f-gumustas)
Results serialization improvements (PR #23909 by @glenn-jocher)
CI maintenance update (PR #24396 by @UltralyticsAssistant)
v3.0.2 to v3.0.3 in workflows.yolo solutions heatmap: documented colormap syntax now behaves correctly out of the box 🎉.imread), cleaner serialization, and several bug fixes reduce runtime surprises.data arg is not passed by @lakshanthad in https://github.com/ultralytics/ultralytics/pull/24383None in imread on read failure to match cv2.imread by @Y-T-G in https://github.com/ultralytics/ultralytics/pull/24406ultralytics 8.4.47 Fix CLI parsing for solution colormap values by @raimbekovm in https://github.com/ultralytics/ultralytics/pull/24219Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.4.46...v8.4.47
Ultralytics v8.4.46 is a stability-focused release that primarily fixes a key multi-scale training edge case 🔧, while also improving export reliabilit
Ultralytics v8.4.46 is a stability-focused release that primarily fixes a key multi-scale training edge case 🔧, while also improving export reliability, hardware support clarity, and documentation quality across YOLO workflows.
🚨 Priority fix (PR #24394 by @glenn-jocher): Multi-scale training minimum size clamp
8.4.45 → 8.4.46.🔁 Resume training safety fix (PR #24386 by @lmycross)
📦 Export usability and correctness improvements
data and fraction for INT8 calibration (PR #24382).🧠 RKNN support clarified and enforced (PR #24384 by @lakshanthad)
rv1103, rv1106, rv1103b, rv1106b) with a helpful error.📚 Docs and content polish
More robust training 🛡️
The multi-scale clamp fix prevents invalid tiny image sizes, reducing crash risk and instability during training—especially for small-image or aggressive augmentation setups.
Less wasted compute/time ⏱️
The resume fix avoids accidental extra epochs when training is already complete.
Cleaner deployment experience 🚀
Better export path reporting and clearer per-format argument support make exporting easier to automate and debug.
Fewer hardware surprises 🤝
RKNN users now get immediate, explicit feedback on unsupported INT8-only chips, avoiding confusing late-stage failures.
Better docs trustworthiness ✅
Improved documentation consistency helps both new and advanced users move faster with fewer mismatches between docs and actual behavior.
data and fraction args across export formats by @lakshanthad in https://github.com/ultralytics/ultralytics/pull/24382ultralytics 8.4.46 Fix multiscale minimum train size by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/24394Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.4.45...v8.4.46
Ultralytics v8.4.45 is a hotfix release that primarily restores correct pretrained training behavior for .pt models (like YOLO26 checkpoints), plus im
Ultralytics v8.4.45 is a hotfix release that primarily restores correct pretrained training behavior for .pt models (like YOLO26 checkpoints), plus important TensorRT/Jetson export reliability improvements and clearer docs. 🚀
🔥 Critical training hotfix (PR #24378 by @glenn-jocher):
YOLO("*.pt").train() and CLI training correctly start from pretrained checkpoint weights.pretrained=False now only disables weight loading when explicitly set.⚙️ TensorRT compatibility hardening:
check_tensorrt() for more consistent setup.🤖 Jetson / INT8 export improvements (PR #24368):
📦 Export argument support expanded:
data as a valid argument for more export formats (including TensorRT, OpenVINO, CoreML, TFLite, TF.js, MNN, IMX), improving dataset-aware INT8/export workflows.📚 Documentation updates:
.pt checkpoints could silently skip pretrained weights, which could reduce accuracy and waste training time.pretrained=False.data argument support reduces validation friction across backends.tensorrt==10.2.0 to fix libnvinfer_builder_resource_win.so.10.2.0 error by @lakshanthad in https://github.com/ultralytics/ultralytics/pull/24367tensorrt from 10.3 to 10.7 on JetPack 6 systems to fix YOLO26 int8 build issues by @lakshanthad in https://github.com/ultralytics/ultralytics/pull/24368data argument to valid args for INT8 export formats by @lakshanthad in https://github.com/ultralytics/ultralytics/pull/24362ultralytics 8.4.44 Apply pretrained arg across model trainers by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/24374ultralytics 8.4.45 Fix pretrained checkpoint training regression by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/24378Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.4.43...v8.4.45
Ultralytics v8.4.44 improves training control and deployment reliability, led by a key fix that makes the pretrained setting behave consistently acros
Ultralytics v8.4.44 improves training control and deployment reliability, led by a key fix that makes the pretrained setting behave consistently across trainers 🔧🚀.
(Top priority) pretrained argument now works consistently in training flows (@glenn-jocher) ✅
pretrained=False is now properly respected even when training from a loaded .pt checkpoint.Export improvements for INT8 workflows 📦
data as a valid export argument for more backends (including TensorRT, OpenVINO, CoreML, TFLite, TF.js, MNN, IMX, and others), reducing “unsupported argument” friction.Jetson + TensorRT stability updates 🤖
end2end auto-disable logic for INT8 export is now narrowly targeted to the known problematic combo (JetPack 6 + TensorRT 10.3.0), instead of broader disabling.TensorRT version guardrails updated 🛡️
10.2.0 (replacing older 10.1.0 exclusion) in loading/export checks.Documentation updates 📚
More predictable training outcomes 🎯
Users can trust that pretrained=False truly starts from random initialization, and pretrained="path/to/weights.pt" is honored correctly.
Easier experiment control and reproducibility 🔁
Reusing checkpoint configs while controlling weight initialization makes transfer learning and ablation experiments cleaner and less error-prone.
Smoother deployment on edge devices ⚡
Jetson/TensorRT fixes reduce export failures, especially for INT8 YOLO26 workflows.
Fewer environment-related surprises 🧩
TensorRT version protections help avoid known broken combinations before they cause runtime/export errors.
Better usability for a broad audience 🌍
Clearer docs and improved Platform dataset tooling (clustering) help both advanced practitioners and newer users work faster with higher confidence.
tensorrt==10.2.0 to fix libnvinfer_builder_resource_win.so.10.2.0 error by @lakshanthad in https://github.com/ultralytics/ultralytics/pull/24367tensorrt from 10.3 to 10.7 on JetPack 6 systems to fix YOLO26 int8 build issues by @lakshanthad in https://github.com/ultralytics/ultralytics/pull/24368data argument to valid args for INT8 export formats by @lakshanthad in https://github.com/ultralytics/ultralytics/pull/24362ultralytics 8.4.44 Apply pretrained arg across model trainers by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/24374Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.4.43...v8.4.44
Ultralytics v8.4.43 mainly improves dataset reliability by fixing how NDJSON and Ultralytics Platform (ul://.../datasets/...) datasets are resolved du
Ultralytics v8.4.43 mainly improves dataset reliability by fixing how NDJSON and Ultralytics Platform (ul://.../datasets/...) datasets are resolved during validation, not just training ✅🚀
Major fix (PR #24365 by @glenn-jocher): unified NDJSON/Platform dataset handling
convert_ndjson_to_yolo_if_needed() in ultralytics.data.utils.ul:// datasets correctly.8.4.42 → 8.4.43.Docs update
Minor cleanup (PR #24351 by @ahmet-f-gumustas)
for ... in dict.keys() loops with direct dict iteration across several files.More consistent dataset behavior across train + val
Users working with NDJSON or Ultralytics Platform datasets now get the same automatic conversion behavior in both stages, reducing surprises. ✅
Fewer validation-time failures
The key practical improvement is preventing validation setup issues when data is provided as .ndjson or ul://... URIs. This makes workflows more robust, especially when running validation independently. 🛡️
Better stability in distributed environments
The conversion step is now coordinated for multi-process runs, helping avoid duplicate or conflicting setup work. 🖥️🖥️
Low upgrade risk
Aside from the dataset-resolution fix, the rest is mostly internal cleanup with no intended user-facing behavior changes. 👍
.keys() calls in dict iteration by @ahmet-f-gumustas in https://github.com/ultralytics/ultralytics/pull/24351ultralytics 8.4.43 Fix NDJSON dataset validation by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/24365Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.4.42...v8.4.43
Ultralytics v8.4.42 focuses on more reliable training under GPU memory pressure 🧠💪, plus several stability, security, export, and documentation improv
Ultralytics v8.4.42 focuses on more reliable training under GPU memory pressure 🧠💪, plus several stability, security, export, and documentation improvements that make YOLO workflows smoother for both developers and end users.
Top priority (current PR #24360 by @glenn-jocher): Better OOM recovery during training 🚀
batch, loss, preds) and trainer loss state are now explicitly cleared before retry.Security hardening in downloads and path handling 🔐
Export pipeline refactor + reliability fixes 📦
Backend usability improvement ✨
model(x) instead of only model.forward(x).Bug fixes affecting real workflows 🛠️
./[data]/).plot_results() crash path when CSV plotting fails.one2many branches are absent.Platform and docs improvements 📚
segments (not polygon).show=False, profiling).1.6.0.Fewer failed training runs on limited VRAM 🎯
The main v8.4.42 change helps training recover more reliably from first-epoch OOM events, reducing frustrating restarts.
Safer defaults for file and dataset handling 🛡️
Security-focused path and extraction checks reduce risk when working with external files and archives.
More dependable exports across formats 🌍
Export refactors and CoreML fixes improve cross-platform conversion stability, especially for deployment-heavy teams.
Better developer ergonomics 👩💻
Callable backends, cleaner export internals, and bug fixes in tuning/plotting/path-loading improve everyday productivity.
Clearer onboarding and learning resources for all users 🙌
Improved docs (including Ultralytics Platform guidance and Jetson memory tips) make setup and adoption easier for newcomers while still useful for advanced users.
__call__ wrapper for BaseBackend.forward method by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/24311ax variable in plot_results by @Y-T-G in https://github.com/ultralytics/ultralytics/pull/24345FileNotFoundError when loading datasets from paths containing square brackets by @prashansapkota in https://github.com/ultralytics/ultralytics/pull/24353one2many head exists while fusing prompt embeddings by @Y-T-G in https://github.com/ultralytics/ultralytics/pull/24359exporter.py into per-format utility modules by @lakshanthad in https://github.com/ultralytics/ultralytics/pull/24245ultralytics 8.4.42 Free tensors before OOM batch retry by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/24360Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.4.41...v8.4.42
Ultralytics v8.4.41 focuses on a key SAM3 video tracking quality fix (fewer ghost IDs) plus a data pipeline reliability improvement for NDJSON dataset
Ultralytics v8.4.41 focuses on a key SAM3 video tracking quality fix (fewer ghost IDs) plus a data pipeline reliability improvement for NDJSON datasets, with a large refresh of docs and Ultralytics Platform guidance 📈✨
🎯 Major tracking fix (current PR, #24249 by @Y-T-G):
init_trk_keep_alive and max_trk_keep_alive from 30 → 10), so stale tracks are removed faster.🗂️ Safer NDJSON dataset conversion (#24290 by @glenn-jocher):
class_names are missing (infers class count more robustly).📚 Documentation and usability upgrades (multiple PRs):
cfg argument (yolo cfg=...) for reusable custom config files.🧪 CI benchmark stability updates (#24286):
ubuntu-latest to cpu-latest.📦 Version bump: 8.4.40 → 8.4.41.
ubuntu-latest to cpu-latest by @lakshanthad in https://github.com/ultralytics/ultralytics/pull/24286ultralytics 8.4.41 Avoid mutable NDJSON dataset cache collisions by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/24290ultralytics 8.4.41 Fix SAM3 FP ghost IDs in video tracking by @Y-T-G in https://github.com/ultralytics/ultralytics/pull/24249Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.4.40...v8.4.41
Ultralytics v8.4.40 introduces per-image precision/recall/F1 tracking during validation (led by PR #24089 from @Laughing-q), making it much easier to
Ultralytics v8.4.40 introduces per-image precision/recall/F1 tracking during validation (led by PR #24089 from @Laughing-q), making it much easier to see exactly which images your model handles well or poorly. 📈🖼️
precision, recall, f1, tp, fp, fn for each image.metrics.box.image_metrics (and also for seg and pose where applicable). ✅image_metrics correctly across ranks, so results remain complete in larger training setups. 🧠⚙️image_metrics storage8.4.39 ➜ 8.4.40 🚀ultralytics 8.4.40 Per-image Precision and Recall by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/24089Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.4.39...v8.4.40
Ultralytics v8.4.39 is a quality-and-usability release focused on clearer run naming (exp-2), better CLI coverage for Solutions, safer rotated-box tra
Ultralytics v8.4.39 is a quality-and-usability release focused on clearer run naming (exp-2), better CLI coverage for Solutions, safer rotated-box training behavior, and broad documentation/platform clarity improvements. 🚀
(Most important) Dashed run path increments are now default via PR #24193 by @glenn-jocher ✅
increment_path() now creates names like exp-2 and results-2.txt (instead of exp2, results2.txt).New CLI parity for Solutions via PR #24251 by @raimbekovm 🧰
yolo solutions regionyolo solutions securityyolo solutions parkingRotated OBB training safety fix via PR #24260 by @lmycross 🛠️
gt_bboxes in RotatedTaskAlignedAssigner by cloning before edits.Small reliability hardening for security alarm flow 🔒
CI workflow resilience improvements via PR #24261 by @glenn-jocher ♻️
Documentation and platform clarity updates 📚
Cleaner experiment tracking 🗂️
The exp-2 style is easier to read and more consistent across files/folders, reducing confusion when managing many runs.
Faster adoption of Solutions from CLI ⚡
Users can now launch region counting, security monitoring, and parking workflows directly from terminal commands with less setup friction.
More stable rotated-box training pipelines 🎯
The OBB fix reduces hidden side effects and hard-to-debug training issues, improving trust in advanced detection workflows.
Better operational reliability 🧪
CI retry hardening helps reduce flaky failures in automation and publishing workflows.
Clearer user guidance and platform onboarding 🌐
Docs now better reflect current best paths (especially Ultralytics Platform) and reduce ambiguity in integrations, licensing, and region details.
gt_bboxes modification in RotatedTaskAlignedAssigner by @lmycross in https://github.com/ultralytics/ultralytics/pull/24260ultralytics 8.4.39 Use dashed incremented run paths by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/24193Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.4.38...v8.4.39
Ultralytics v8.4.38 is a stability-focused release that mainly improves model export reliability and consistency across many deployment formats, with
Ultralytics v8.4.38 is a stability-focused release that mainly improves model export reliability and consistency across many deployment formats, with additional fixes for training, tracking, and SAM3 behavior. 🚀
"image"), improving compatibility.output_file, output_dir, etc.).lrpc is present.python -m pip (works better in restricted environments).rknn-toolkit2>=2.3.2).--pre) where needed.track_buffer now consistently behaves as a true frame count in ByteTrack/BoT-SORT (no hidden FPS scaling).track_buffer now matches what you configure, reducing surprising track drop behavior at non-30 FPS. 🎥torch and torchvision versions on JetPack 5 Docker by @lakshanthad in https://github.com/ultralytics/ultralytics/pull/24199test_engine.py and add OBB/Pose test coverage by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/24197trainer.stride assignment for DDP training by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/24208annotation.md by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/24232axelera-runtime installation command to allow prerelease versions by @lakshanthad in https://github.com/ultralytics/ultralytics/pull/24230lrpc attribute before attempting fuse during YOLOE model export by @Y-T-G in https://github.com/ultralytics/ultralytics/pull/24239fuse method by @Y-T-G in https://github.com/ultralytics/ultralytics/pull/24246ultralytics 8.4.38 Unify args naming for standalone export functions by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/24120Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.4.37...v8.4.38
Ultralytics v8.4.37 is a quality + workflow-focused release: the tag PR itself is a version bump, while the main substance is improved hyperparameter
Ultralytics v8.4.37 is a quality + workflow-focused release: the tag PR itself is a version bump, while the main substance is improved hyperparameter tuning (now NDJSON-based for multi-dataset runs), better handling of class imbalance, stronger training reliability, and clearer docs/UI guidance. 🚀
[!WARNING] The mAP calculation has been revised in this release. Reported mAP may be slightly lower than in previous Ultralytics versions, but now more closely matches pycocotools' COCOEval.
(Priority note) Current PR #24192 by @glenn-jocher: release tag/version update only (8.4.36 → 8.4.37) 📦
Major tuning upgrade (PR #24179 by @Laughing-q) 🧠
tune_results.ndjson.tune_fitness.png), and MongoDB sync now aligns with local NDJSON logs.Class imbalance support in training (PR #23565 by @ahmet-f-gumustas) ⚖️
cls_pw hyperparameter to weight underrepresented classes more during detection training.0.0), so existing behavior stays unchanged unless enabled.Training stability + robustness fixes 🛡️
safe_download path handling) (PR #24185 by @glenn-jocher).Evaluation and CI reliability improvements ✅
compute_ap precision edge-case fix for more robust AP calculation (PR #24175 by @Laughing-q).Cleaner distributed training logs (PR #24177 by @Laughing-q) 🧹
Documentation and platform UX improvements 📚
.load() weights in segment/OBB docs.cls_pw can improve rare-class learning without changing your whole pipeline. 🎯compute_ap by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/24175Dockerfile-nvidia-arm64 by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/24181ultralytics 8.4.37 NDJSON-based multidataset hyperparameter tuning by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/24179ultralytics 8.4.37 NDJSON-based multidataset hyperparameter tuning by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/24192Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.4.36...v8.4.37
Ultralytics v8.4.36 is a stability-focused release that fixes an important training regression for checkpoint-based workflows (especially Ultralytics
Ultralytics v8.4.36 is a stability-focused release that fixes an important training regression for checkpoint-based workflows (especially Ultralytics Platform/HUB usage), plus several documentation clarifications for Platform, Explorer, and Jetson guides. ✅
🧠 Critical training fix in Model.train() (PR #24167 by @glenn-jocher)
.pt model is already loaded, training reuses that loaded model directly.trainer.args.model (which may be a non-file identifier in Platform/HUB flows).🧪 Regression coverage added
📚 Platform and docs alignment updates (supporting PRs)
ultralytics>=8.4.35 for integration.>=8.3.12, with guidance to pin 8.3.11 if needed.ultralytics 8.4.36 Fix platform training checkpoint regression by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/24167Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.4.35...v8.4.36
Ultralytics v8.4.35 is a stability-focused release that makes training recovery smarter, dataset caching safer, and inference/runtime behavior more re
Ultralytics v8.4.35 is a stability-focused release that makes training recovery smarter, dataset caching safer, and inference/runtime behavior more reliable—especially when runs hit NaNs or dataset metadata is inconsistent. 🚀🛡️
NaN training recovery improved (most important, PR #24154 by @glenn-jocher) 🔁
last_good.pt instead of retrying a potentially corrupted last.pt.Detection dataset cache reliability upgrades (PR #24154) 🗂️
Cleaner logs during runs (PR #24154) 🔕
Platform: Model not found warnings are throttled to reduce noise.OpenVINO robustness improvements (PR #24156 by @glenn-jocher) ⚙️
Training behavior and compatibility fixes
isatty() in console capture wrapper to avoid crashes with some libraries like transformers (PR #24143).Docs and ecosystem updates 📚
Fewer failed long trainings ⏱️
If your run encounters NaNs, recovery is now more trustworthy, reducing wasted GPU time and repeated crash loops.
Less dataset-debug frustration 🧪
Better cache validation + clearer corrupt-label messages make it faster to diagnose data issues.
More predictable production behavior 🏭
OpenVINO and logging adjustments improve runtime stability on edge/ARM setups and reduce noisy warnings.
Smoother experiment tracking and resuming 📈
W&B resume continuity helps keep metrics in one place, improving experiment history quality.
Low-risk upgrade with practical benefits ✅
This release is mostly about reliability and developer experience rather than new model architectures—great for teams running frequent training/inference pipelines.
segment2box and scale_boxes in utils/ops.py by @ahmet-f-gumustas in https://github.com/ultralytics/ultralytics/pull/24127YAML.dump usage in benchmark.py by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/24146sahi tiled inference code by @RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/24133isatty() to _ConsoleCapture stream wrapper by @fcakyon in https://github.com/ultralytics/ultralytics/pull/24143Model.train() for YAML models by @artest08 in https://github.com/ultralytics/ultralytics/pull/23640ultralytics 8.4.35 NaN training recovery and dataset cache improvements by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/24154Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.4.34...v8.4.35
Ultralytics v8.4.34 is a tuning and stability-focused release 🚀, led by a major new feature: multi-dataset hyperparameter tuning in one run, plus seve
Ultralytics v8.4.34 is a tuning and stability-focused release 🚀, led by a major new feature: multi-dataset hyperparameter tuning in one run, plus several important reliability fixes and broad YOLO26 documentation updates.
🧠 Major feature (PR #24067 by @Laughing-q): Multi-dataset hyperparameter tuning
model.tune() now accepts data as either a single dataset or a list.coco8.yaml + coco8-grayscale.yaml).8.4.34.🛡️ Training resume stability fix (PR #24085 by @Y-T-G)
exp_avg_sq state in FP32.🔒 Thread-safe ONNX export (PR #24092 by @glenn-jocher)
⚙️ Robustness fixes in core runtime
AAttn fixed for non-divisible dim/num_heads cases to avoid shape/group crashes (PR #24114 by @ZoomZoneZero).crop_mask() now clamps negative coordinates before cropping for safer segmentation postprocessing (PR #24115 by @Y-T-G).draw_specific_kpts() now respects user-provided keypoint index order and handles missing confidence values safely (PR #24099 by @onuralpszr).📚 Documentation and ecosystem refresh (many PRs)
cuDSS dependency instructions for Torch 2.10.0 (PR #24081 by @lakshanthad).Better real-world tuning quality 🎯
Multi-dataset tuning helps teams optimize one model for mixed or varied data domains (for example, color + grayscale, or multiple data sources), improving generalization and reducing overfitting to a single dataset.
More reliable training and export workflows ✅
Resume training is more stable, distributed cleanup is safer, and ONNX export is more dependable in threaded environments—especially useful in production pipelines.
Improved deployment and edge guidance 📱
Updated YOLO26 Jetson benchmarks and setup docs make edge deployment decisions more current and practical.
Cleaner user experience in docs and platform onboarding ✨
Better Smart Annotation and dataset split guidance can reduce setup friction and speed up annotation/training workflows on the Ultralytics Platform.
cuDSS package to JetPack 6 installation guide by @lakshanthad in https://github.com/ultralytics/ultralytics/pull/24081draw_specific_kpts to respect user-specified indices order by @onuralpszr in https://github.com/ultralytics/ultralytics/pull/24099crop_mask before cropping by @Y-T-G in https://github.com/ultralytics/ultralytics/pull/24115num_heads by @ZoomZoneZero in https://github.com/ultralytics/ultralytics/pull/24114ultralytics 8.4.34 Multi-dataset support for hyperparameter tuning by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/24067Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.4.33...v8.4.34
Ultralytics v8.4.33 focuses on a key training reliability fix for end-to-end YOLO workflows, plus improvements to Ray Tune flexibility and CoreML expo
Ultralytics v8.4.33 focuses on a key training reliability fix for end-to-end YOLO workflows, plus improvements to Ray Tune flexibility and CoreML export stability 🚀
🔧 Major fix (current PR #24074 by @Laughing-q): Resume-training for end-to-end models now restores loss state correctly
resume_training() to properly reinitialize and sync the model’s loss criterion when loading checkpoints.8.4.32 to 8.4.33.🧠 Ray Tune upgrades (PR #23946 by @lmycross): More search algorithms supported in YOLO26 tuning
search_alg to model.tune(..., use_ray=True) with options like optuna, hyperopt, bohb, ax, nevergrad, zoopt, random, and more.iterations (clearer API behavior).HyperBandForBOHB) instead of generic defaults.close_mosaic now uses integer sampling).🍎 CoreML export fix (PR #24078 by @glenn-jocher): Better detection export with NMS
nms=True.More reliable resumed training ✅
If training is interrupted and resumed from a checkpoint, end-to-end models now continue with the correct loss progression instead of partially reset behavior. This is the most important user-facing fix in this release.
Stronger hyperparameter tuning workflows 🎯
YOLO26 users get more control over search strategy in Ray Tune, making large-scale tuning more adaptable to different infrastructure and optimization preferences.
Smoother mobile/Apple deployment 📱
CoreML detection exports with integrated NMS are more robust, lowering chances of deployment-time surprises.
Bottom line: v8.4.33 is a stability + flexibility release—especially valuable for users resuming long trainings and teams doing advanced automated tuning.
ultralytics 8.4.33 Progressive loss train resume fix by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/24074Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.4.32...v8.4.33
v8.4.32 is mainly an Axelera AI export expansion release 🚀—it significantly improves how Ultralytics models (including more tasks) are exported and de
v8.4.32 is mainly an Axelera AI export expansion release 🚀—it significantly improves how Ultralytics models (including more tasks) are exported and deployed on Axelera hardware, with supporting docs and usability updates across the Ultralytics ecosystem.
Major (Current PR #23844): Axelera export pipeline refactor and expansion 🧠⚙️
ultralytics/utils/export/axelera.py.torch2axelera (replacing older ONNX-centered flow), making the pipeline cleaner and easier to maintain.TASK2CALIBRATIONDATA, so exports pick better default calibration datasets automatically.Axelera runtime/backend improvements 🔌
axelera-rt / axelera-devkit 1.6.0rc3 path)..axm execution pipeline more directly.New practical examples for deployment 📁🎯
Documentation upgrades (large part of this release) 📚
.tar.gz / .tgz and NDJSON.For developers:
For edge deployment users (Axelera):
For Ultralytics Platform users:
Overall:
ultralytics 8.4.32 Expand Axelera exportable models and tasks by @lakshanthad in https://github.com/ultralytics/ultralytics/pull/23844Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.4.31...v8.4.32
Ultralytics v8.4.31 is a reliability-focused release that mainly fixes INT8 export calibration for non-square image sizes (the headline change), while
Ultralytics v8.4.31 is a reliability-focused release that mainly fixes INT8 export calibration for non-square image sizes (the headline change), while also improving training stability, export maintainability, and documentation for deployment and dataset workflows 🚀
🔥 Main update (PR #24028 by @Y-T-G): INT8 calibration now works correctly with non-square imgsz
imgsz=640,480 with int8.LetterBox behavior fixed).🧩 Export system refactor (PR #23914 by @onuralpszr)
torchscript, openvino, coreml, ncnn, mnn, paddle, rknn, axelera, etc.).🍎 Apple Silicon stability improvement (PR #24038 by @Y-T-G)
📦 Better auto-batch with multi-scale (PR #24051 by @glenn-jocher)
multi_scale is enabled.📚 Docs and usability upgrades
🛠️ CI and environment robustness
ubuntu-latest and Codecov v6 updates.--no-tty).✅ Dataset conversion validation tightening (PR #24031 by @glenn-jocher)
val split (instead of allowing test as substitute).--no-tty flag for edgetpu_compiler installation by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/24032Exporter export methods into per-format utility modules by @onuralpszr in https://github.com/ultralytics/ultralytics/pull/23914ubuntu-latest by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/24049multi_scale in auto-batch computation by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/24051ultralytics 8.4.31 INT8 calibration with non-square imgsz by @Y-T-G in https://github.com/ultralytics/ultralytics/pull/24028Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.4.30...v8.4.31
Ultralytics v8.4.30 is a focused stability release that fixes and hardens training resume behavior, making interrupted training runs much more reliabl
Ultralytics v8.4.30 is a focused stability release that fixes and hardens training resume behavior, making interrupted training runs much more reliable 🔄✅.
trainer.py to correctly restore training arguments from last.pt earlier in the resume flow.data argument instead of failing unexpectedly.imgsz, batch, device, workers, cache, freeze, val, and plots (plus related options).augmentations again when resuming, since they can’t be auto-restored exactly.8.4.29 ➜ 8.4.30.ultralytics 8.4.30 Fix training resume by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/24027Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.4.29...v8.4.30
Ultralytics v8.4.29 is mainly a training reliability release 🔧—it makes resume=True much safer and clearer, while also adding a helpful new COCO JSON
Ultralytics v8.4.29 is mainly a training reliability release 🔧—it makes resume=True much safer and clearer, while also adding a helpful new COCO JSON training guide and small CI/docs/test maintenance updates.
Major (current PR #24021 by @glenn-jocher): safer training resume flow 🔄✅
resume=True now only resumes if the checkpoint truly contains resumable training state (like epoch + optimizer state).imgsz, batch, device, workers, cache, patience, validation flags, etc.) ⚙️.8.4.28 to 8.4.29 📦.New docs feature (PR #24005 + cleanup in #24022): train directly from COCO JSON 📘
.txt labels.CI benchmark alignment with latest model family (PR #23965) 🚀
Stability and maintenance 🧪
zidane.jpg.onnx, requests).ultralytics 8.4.29 Training resume fixes by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/24021Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.4.28...v8.4.29
Ultralytics v8.4.28 improves training reliability on small datasets by making autobatch smarter, while also strengthening network robustness, backend
Ultralytics v8.4.28 improves training reliability on small datasets by making autobatch smarter, while also strengthening network robustness, backend efficiency, and docs/test stability. 🚀
(Most important) Autobatch now respects dataset size 📦
From PR #24020 by @glenn-jocher: automatic batch-size selection is now capped to the number of training images.
dataset_size through Trainer.auto_batch(), check_train_batch_size(), and autobatch().More reliable dataset/platform networking 🌐🔁
From PR #24010 by @glenn-jocher:
TensorFlow backend import optimization ⚡
From PR #24017 by @Y-T-G:
saved_model and pb formats in the TensorFlow backend loader.Test reliability improvement for YOLO26 OBB CI 🧪
From PR #24011 by @glenn-jocher:
boats.jpg) to a fixed jsDelivr CDN path for more consistent test behavior.Ultralytics Platform docs updated to match current UI 📘
From PR #24014 by @mykolaxboiko:
Better small-dataset training defaults ✅
Users training on tiny datasets are less likely to get impractical batch recommendations, reducing setup friction and odd edge-case behavior.
More resilient cloud/data workflows 🛡️
Dataset downloads and platform URL resolution should fail less often on temporary network issues, especially in CI or unstable network environments.
Cleaner runtime behavior 🧹
Conditional TensorFlow imports can reduce startup overhead and avoid avoidable environment-related issues.
Improved developer experience 👩💻
More stable tests and clearer docs mean smoother onboarding, troubleshooting, and release confidence.
ultralytics 8.4.28 Limit autobatch to dataset size by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/24020Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.4.27...v8.4.28
Ultralytics v8.4.27 focuses on more reliable Ultralytics Platform training control (especially cancellation handling) ✅, plus several stability fixes
Ultralytics v8.4.27 focuses on more reliable Ultralytics Platform training control (especially cancellation handling) ✅, plus several stability fixes for data conversion, postprocessing alignment, Paddle compatibility, and Docker runtime updates 🚀.
🔴 Priority update (PR #24008 by @glenn-jocher): Improved Platform training integration
🧩 COCO conversion robustness (PR #23998 by @raimbekovm)
convert_coco() crashes when annotations are missing or have empty keypoints/segmentation.IndexError.🎯 Coordinate/mask alignment fix (PR #23995 by @raimbekovm)
scale_coords() and scale_masks() to match LetterBox behavior.🐼 Paddle dependency safety pin (PR #23997 by @Laughing-q)
<3.3.0 for export/inference paths to avoid known breakages.🐳 Docker runtime refresh (PR #23991 by @glenn-jocher)
2.11.0 (CUDA 12.8, cuDNN 9 unchanged).📚 Documentation discoverability (PR #23980 by @raimbekovm)
More dependable Platform workflows 🛡️
Fewer training tracking failures and more consistent cancellation behavior in cloud-managed runs on the Ultralytics Platform.
Better training control UX ⏹️
Cancel signals are handled more predictably, reducing confusion and wasted compute.
Fewer dataset conversion interruptions 📦
Users converting imperfect COCO labels are less likely to hit hard crashes.
Cleaner prediction geometry 🖼️
Improved coordinate/mask consistency helps downstream quality for segmentation and pose tasks.
Safer dependency behavior ⚙️
Paddle users are protected from problematic 3.3.x versions by default.
Modernized container baseline 🚀
Docker users get a newer PyTorch runtime with minimal GPU stack risk.
paddlepaddle<3.3.0 by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/23997scale_coords and scale_masks to match LetterBox by @raimbekovm in https://github.com/ultralytics/ultralytics/pull/23995ultralytics 8.4.27 Improved Platform training integration by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/24008Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.4.26...v8.4.27
v8.4.26 focuses on reliability and usability improvements: smarter Platform dataset handling (auto-validation split for NDJSON), more robust Platform
v8.4.26 focuses on reliability and usability improvements: smarter Platform dataset handling (auto-validation split for NDJSON), more robust Platform URL resolution, and an important FP16 SAM inference crash fix—plus CI and docs polish. 🚀
✅ Platform NDJSON auto-split (PR #23990, @glenn-jocher)
If a dataset has a train split but no val/test, Ultralytics now automatically creates a small validation split from training data instead of failing immediately.
🌐 More reliable Ultralytics Platform URI resolution (PR #23990, @glenn-jocher)
Improved ul://... resolution with:
🧠 FP16 SAM TinyViT inference crash fix (PR #23780, @Edwin-Kevin)
Fixed a half-precision inference error in SAM TinyViT models (like mobile_sam.pt) caused by dtype mismatch in cached tensors.
half=True.⚙️ CI simplification and stability updates (mainly PR #23990)
cpu-latest to ubuntu-latesttorch<2.11 until CUDA 13 driver support is ready📚 Platform docs improvements
v0.2.7 via PR #23989 by @glenn-jocherLess friction for dataset conversion 📦
Users importing NDJSON datasets into YOLO workflows are less likely to hit hard failures when a validation split is missing.
Better reliability in cloud/remote workflows ☁️
Platform URI retries and timeout tuning reduce flaky failures, especially with large datasets or temporary network instability.
More stable FP16 segmentation inference 🎯
Users running SAM TinyViT in half precision on GPU should see fewer runtime dtype crashes and smoother deployment behavior.
Cleaner maintenance pipeline for Ultralytics 🛠️
CI changes reduce complexity and improve consistency, helping keep releases stable.
Faster onboarding for new users 🎥
The new Platform tutorial videos make it easier for broad audiences to get started quickly with data, training, and deployment.
ultralytics 8.4.26 Platform NDJSON autosplit by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/23990Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.4.25...v8.4.26
Ultralytics v8.4.25 mainly restores TensorFlow.js export for most users ✅, while also improving training speed paths, visualization behavior, box alig
Ultralytics v8.4.25 mainly restores TensorFlow.js export for most users ✅, while also improving training speed paths, visualization behavior, box alignment accuracy, CI reliability, and YOLO26/platform documentation clarity 🚀📚.
(Top priority) TF.js export restored for supported systems 🔧
From PR #23985 by @glenn-jocher:
ydf<0.13.0 (non-ARM64) to avoid a TensorFlow/protobuf conflict.8.4.25.Faster training internals for pose/detection/OBB losses ⚡
From PRs #23937 and #23966:
Bounding box alignment fix in scale_boxes 🎯
From PR #23967:
LetterBox rounding.Visualization improvement: confidence can be shown without class labels 👀
From PR #23951:
show_conf=True now works independently of show_labels.Docs and platform guidance expanded (especially YOLO26 end-to-end detection) 📘
From PRs #23947, #23974, #23975, #23931:
References sections).CI and runner infrastructure updates 🧪
From PR #23875 and #23974:
cpu-latest for better consistency/speed.2.333.0.Prediction telemetry enhancement 📡
From PR #23948:
Big win for web deployment 🌐
Teams exporting to TF.js can now resume browser/JavaScript workflows in most supported environments.
Fewer export headaches 🧩
Dependency pinning reduces breakage from upstream package mismatches, making exports more dependable.
Potentially faster training iterations ⏱️
Vectorized loss preprocessing should reduce Python overhead and improve throughput, especially for larger batches.
Better output quality and trust in predictions ✅
The box-scaling fix helps prevent annoying small coordinate offsets that can affect visual quality and downstream logic.
Improved usability for debugging and demos 🛠️
Confidence-only display and richer backend telemetry make model behavior easier to inspect.
Smoother onboarding for YOLO26 + Platform users 🚀
Updated docs lower migration friction and make it easier for both new and advanced users to train/deploy via the Ultralytics Platform.
cpu-latest runner for faster CI tests by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/23875detect/obb loss calculation by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/23966References docstring sections rendering incorrectly in API docs by @raimbekovm in https://github.com/ultralytics/ultralytics/pull/23931scale_boxes to match LetterBox rounding by @a17sol in https://github.com/ultralytics/ultralytics/pull/23967ultralytics 8.4.25 TensorFlow.js fix pin ydf<0.13.0 by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/23985Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.4.24...v8.4.25
Ultralytics `v8.4.24` improves training reliability and clarity on Ultralytics Platform 🎯, while also aligning tuning defaults and docs with YOLO26 be
Ultralytics v8.4.24 improves training reliability and clarity on Ultralytics Platform 🎯, while also aligning tuning defaults and docs with YOLO26 best practices 🚀.
(Most important) Better Platform training error surfacing 🛠️
Version bump to 8.4.24 🔖
8.4.23 to 8.4.24.Ray Tune search space updated to match modern YOLO26 ranges 📈
lr0, momentum, box, cls, scale).dfl and close_mosaic.YOLO26 naming adopted in Streamlit inference selector 🔄
yolo11* to yolo26* in the Streamlit solution UI.Rockchip RKNN docs refreshed with YOLO26 benchmarks 📘
TF.js benchmarking disabled (temporary safeguard) ⚠️
Platform docs and UX docs improvements 🧾
Faster troubleshooting for Platform users ✅
More precise error messages reduce guesswork and support time when training jobs fail.
Cleaner failure behavior 🧹
Disabling callbacks after registration failure prevents repeated warning spam and gives a more stable user experience.
Better model tuning outcomes 🎯
Updated Ray Tune ranges help users avoid outdated search spaces that could hurt accuracy, especially with YOLO26 workflows.
Stronger consistency across product and docs 📚
YOLO26 naming, benchmark updates, and docs alignment make it easier for users to follow current recommended paths.
Safer defaults for problematic integrations 🛡️
TF.js benchmark disablement avoids known dependency traps until upstream compatibility improves.
UTM enabled banner redirect by @RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/23952YOLO11 → YOLO26 by @RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/23954ultralytics 8.4.24 Improve Platform train error surfacing by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/23957Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.4.23...v8.4.24
Ultralytics v8.4.23 is mainly a big under-the-hood inference upgrade 🚀: AutoBackend was fully redesigned into modular backend classes, making multi-fo
Ultralytics v8.4.23 is mainly a big under-the-hood inference upgrade 🚀: AutoBackend was fully redesigned into modular backend classes, making multi-format model deployment cleaner, easier to maintain, and more reliable across platforms.
Major architecture refactor (PR #23790 by @Laughing-q) 🧩
AutoBackend moved from one large file to a modular backend system (ultralytics/nn/backends/).BaseBackend interface to standardize model loading, metadata handling, and forward inference behavior.AutoBackend now uses a single format route + backend map instead of many backend boolean flags.channels, format, warmup paths, etc.).docs/en/reference/nn/backends/.Pose dataset conversion improvement (PR #23921 by @glenn-jocher) 🤸
kpt_shape automatically (when possible), instead of failing immediately.Docs correctness and usability fixes 📚
cfg.md defaults for plots and save to True (PR #23917).Docs/site maintenance and CI updates 🛠️
print() to centralized LOGGER in docs reference builder (PR #23895).For users deploying models in different formats 🌍
This release should improve consistency and stability of inference behavior across backends, with fewer backend-specific surprises.
For developers and contributors 👩💻
The modular backend design is a big win: easier to extend, debug, and maintain than a monolithic AutoBackend.
For pose dataset workflows 📦
NDJSON conversion is more forgiving and practical, reducing friction when metadata is incomplete.
For documentation trustworthiness ✅
Multiple fixes make docs more accurate and navigable, reducing confusion around defaults and TensorRT calibration behavior.
Overall, v8.4.23 is a foundational engineering release: not a flashy model change, but a high-impact internal upgrade that strengthens the reliability of the Ultralytics inference stack over time.
ultralytics 8.4.23 Refactor AutoBackend into modular per-backend classes by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/23790Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.4.22...v8.4.23
Ultralytics v8.4.22 focuses on better hardware support and reliability, led by a key new feature: basic Huawei Ascend NPU device support in select_dev
Ultralytics v8.4.22 focuses on better hardware support and reliability, led by a key new feature: basic Huawei Ascend NPU device support in select_device 🚀.
🔥 Major (Current PR #23902 by @GiantAxeWhy): Huawei Ascend NPU parsing added
device=npu or device=npu:0 in Ultralytics.torch_npu installation and NPU availability.npu:0,1 for now.🧠 Multi-GPU training fix (DDP + Albumentations)
📦 Export stability improvements
end2end for incompatible TensorRT versions (<=10.3.0) to prevent known build failures.🐳 Docker and CI environment updates
UV_BREAK_SYSTEM_PACKAGES=1 across Dockerfiles for more consistent installs.Dockerfile-runner-cpu and broader CI runner improvements (CPU labels, tooling like sudo, nodejs, npm, gpg).🛠️ Reliability/bug fixes
📚 Docs and dataset additions
callable type hint to Callable in BasePredictor.add_callback() by @ahmet-f-gumustas in https://github.com/ultralytics/ultralytics/pull/23836task attribute from metrics classes by @raimbekovm in https://github.com/ultralytics/ultralytics/pull/238352026.0.0 from Conda CI by @Y-T-G in https://github.com/ultralytics/ultralytics/pull/23839end2end for TensorRT INT8 exports with tensorrt<=10.3.0 by @lakshanthad in https://github.com/ultralytics/ultralytics/pull/23872gpg package for Dockerfile-runner-cpu by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/23878dota128 dataset by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/23885ModuleNotFoundError in classify caching check to prevent numpy version conflict by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/23892export-table.md by @Y-T-G in https://github.com/ultralytics/ultralytics/pull/23898batch fix by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/23899black dep and fix publish commit message by @raimbekovm in https://github.com/ultralytics/ultralytics/pull/23896cpu-latest by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/23901cpu-latest runner by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/23904ultralytics 8.4.22 Huawei Ascend NPU device parsing in select_device by @GiantAxeWhy in https://github.com/ultralytics/ultralytics/pull/23902Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.4.21...v8.4.22
Ultralytics v8.4.21 improves reliability for Rockchip RKNN exports (main change) and also adds better tuning isolation, expanded C++ pose example supp
Ultralytics v8.4.21 improves reliability for Rockchip RKNN exports (main change) and also adds better tuning isolation, expanded C++ pose example support, and clearer YOLO26 optimizer guidance. 🚀
✅ Main priority (PR #23806 by @Laughing-q): RKNN export path fix
Path(...).stem) instead of fragile string replacement.🧪 Ray Tune reliability improvement (PR #23793 by @Y-T-G)
trainer = None) before running.🧍 ONNXRuntime C++ example gains pose support for YOLOv8-family pose models (PR #23786 by @chendao12138)
📘 YOLO26 training docs improved around MuSGD (PRs #23800, #23804 by @deriiinjv and @monkeyjack123)
optimizer=auto.For deployment users on Rockchip 📦
RKNN exports should now fail less often due to naming/path edge cases, making deployment workflows more dependable.
For ML engineers tuning models 🔁
Ray Tune trials are more isolated and reproducible, reducing hard-to-debug inconsistencies.
For C++/edge developers ⚙️
Easier starting point for pose inference in ONNXRuntime C++ projects using YOLOv8-style pose models.
For YOLO26 trainers 🧠
Optimizer behavior is easier to understand, helping users make better training choices faster.
ultralytics 8.4.21 Fix Rockchip RKNN export path by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/23806Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.4.20...v8.4.21
Ultralytics v8.4.20 is a stability-and-usability release focused on cleaner hyperparameter tuning, more reliable deployment/export workflows, and impr
Ultralytics v8.4.20 is a stability-and-usability release focused on cleaner hyperparameter tuning, more reliable deployment/export workflows, and improved docs for both YOLO models and the Ultralytics Platform 🚀
(Most important) Ray Tune cleanup in current PR #23772 🧹
tuner_callbacks and dropped built-in W&B callback wiring from RunConfig in tuning.8.4.19 → 8.4.20.RKNN export reliability improved (#23802) 📦
onnx<1.19.0) and enforced ONNX opset cap (<=19) for RKNN conversion.FastSAM prompt accuracy fix (#23766) 🎯
ByteTracker consistency update (#23771) 🛠️
fuse_score=True), aligning behavior across tracking stages.Better YAML error messages (#23767) ✅
Jetson/JetPack 6 stack refresh (#23788 + #23801) 🤖
torch 2.10 + torchvision 0.25.Docs and platform improvements 📚
More robust tuning workflows ⚙️
Fewer deployment surprises 🚀
Higher output quality in segmentation/tracking 👀
Smoother setup for modern environments 🔧
Better onboarding and reproducibility 🧠
sam-3.md speed comparison by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/23782torch 2.10 by @lakshanthad in https://github.com/ultralytics/ultralytics/pull/23788torch and torchvision compatibility table to latest by @lakshanthad in https://github.com/ultralytics/ultralytics/pull/23801ultralytics 8.4.20 Remove redundant hardcoded tuner_callbacks by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/23772Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.4.19...v8.4.20
Ultralytics v8.4.19 focuses on much more reliable Ultralytics Platform training sync (especially model tracking via model_id), plus a few quality impr
Ultralytics v8.4.19 focuses on much more reliable Ultralytics Platform training sync (especially model tracking via model_id), plus a few quality improvements for SAM outputs, lightweight model stability, and clearer YOLO26 end-to-end docs 🚀
🔴 Most important (current PR #23761 by @glenn-jocher): Platform training model_id fix
model_id during model uploads and training events.trainer.platform context (instead of scattered trainer fields), including:
model_id🧠 SAM prediction cleanup (#23751 by @Laughing-q)
SAM3SemanticPredictor postprocessing and feature-inference paths.🛡️ PSA attention edge-case fix (#23758 by @Y-T-G)
PSABlock for very small channel configs.📘 Docs clarification for end-to-end models (#23720 by @raimbekovm)
max_det and agnostic_nms are supported directly.end2end=False is mainly for enabling traditional iou-based NMS behavior.⚙️ CI maintenance
For Ultralytics Platform users: ✅
Training runs should now be more reliable to track end-to-end, with fewer mismatches between uploaded checkpoints and the correct model session. This is the biggest practical improvement in this release.
For production and team workflows: 📡
Better callback state management means cleaner event handling, better cancellation behavior, and improved confidence that results and artifacts land in the right place.
For segmentation users (SAM): 🎯
Cleaner predictions with fewer duplicate overlaps can improve downstream usability and visual quality.
For custom/small models: 🧱
Fewer architecture edge-case failures when using very small widths or low channel counts.
For all users reading docs: 📚
Clearer expectations around end-to-end inference/validation args in YOLO26 and YOLOv10 reduce confusion and setup mistakes.
max_det and agnostic_nms support in end2end mode by @raimbekovm in https://github.com/ultralytics/ultralytics/pull/23720SAM3SemanticPredictor to eliminate overlapping boxes by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/23751num_heads by @Y-T-G in https://github.com/ultralytics/ultralytics/pull/23758ultralytics 8.4.19 Platform training pass model ID fix by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/23761Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.4.18...v8.4.19
Object tracking notebook updated from deprecated Annotator.seg_bbox() to modern Results.plot(color_mode="instance").
Ultralytics v8.4.18 is a reliability-focused release that improves downloads and dependency installs, while also adding stronger ExecuTorch export support (including Pose) for smoother edge/mobile deployment 🚀.
safe_download() now handles URLs with spaces by encoding spaces as %20 (PR #23736 by @glenn-jocher).
ultralytics/utils/export/executorch.py with reusable torch2executorch() and executorch_wrapper().git+ requirements with version constraints (PR #23737 by @fcakyon).
Annotator.seg_bbox() to modern Results.plot(color_mode="instance").safe_download() fix is small but high-impact: fewer random failures when fetching models/assets from imperfect URLs.executorch.py by @lakshanthad in https://github.com/ultralytics/ultralytics/pull/23734Annotator.seg_bbox in object tracking notebook by @ahmet-f-gumustas in https://github.com/ultralytics/ultralytics/pull/23745ultralytics 8.4.18 safe_download() URLs with spaces by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/23736Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.4.17...v8.4.18
Ultralytics v8.4.17 makes NDJSON dataset conversions *resplit-friendly*—reusing existing images, cleaning stale labels, and avoiding unnecessary downl
Ultralytics v8.4.17 makes NDJSON dataset conversions resplit-friendly—reusing existing images, cleaning stale labels, and avoiding unnecessary downloads for faster iteration 🚀📦
train/val/test splits.labels/ directory before reconversion to prevent stale annotations.data.yaml to skip reconversion when nothing meaningful changed 🧾🔒end2end mode for EdgeTPU exports (and logs a warning), aligning EdgeTPU with other limited backends that don’t support required ops.nncf requirements to reduce install/export conflicts across Torch versions.end2end behavior, and OpenVINO INT8 export setups are smoother across PyTorch versions.nncf<3 for PyTorch 2.2 and below by @Y-T-G in https://github.com/ultralytics/ultralytics/pull/23726end2end for EdgeTPU by @Y-T-G in https://github.com/ultralytics/ultralytics/pull/23724ultralytics 8.4.17 NDJSON dataset re-split support by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/23735Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.4.16...v8.4.17
Ultralytics 8.4.16 mainly improves Windows ↔ Linux/macOS `.pt` model portability by fixing pathlib pickle issues, with a small docs update for HEIC/HE
Ultralytics 8.4.16 mainly improves Windows ↔ Linux/macOS .pt model portability by fixing pathlib pickle issues, with a small docs update for HEIC/HEIF image support 🧩💾📸
.pt loading fix (PR #23725 by @glenn-jocher):
torch_safe_load() to remap pathlib.WindowsPath ↔ pathlib.PosixPath during unpickling, preventing common load failures when models are created on one OS and loaded on another.model.pt_path is now stored as a plain string (str(weight)), not a Path object (applies in general checkpoint loading and YOLO NAS model wrapper code).8.4.15 → 8.4.16.pi-heif (auto-installed on first use) and AVIF is supported natively by Pillow..heif is explicitly listed among supported image formats..pt weights between teammates or CI systems running different operating systems.model.pt_path (strings behave consistently across platforms and serialization)..heif support visibility (less dependency confusion).ultralytics 8.4.16 Windows<>Linux model compatibility fix by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/23725Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.4.15...v8.4.16
Ultralytics v8.4.15 improves real-world image ingestion (adds HEIC/HEIF + HEIF support with lazy decoding + better EXIF handling) and includes several
Ultralytics v8.4.15 improves real-world image ingestion (adds HEIC/HEIF + HEIF support with lazy decoding + better EXIF handling) and includes several quality fixes for metrics, exports, and docs 🖼️⚙️✅
ultralytics.utils.patches.image_open and monkey-patched PIL.Image.open to lazily enable HEIC/HEIF decoding via pi-heif only when needed (i.e., on first decode failure) ⏱️.heif (in addition to .heic) ✅.ttf fonts from the Ultralytics user config directory for stronger multilingual rendering.ImageOps.exif_transpose() when loading images from lists of paths/URLs, reducing “sideways phone photo” issues.match_predictions) to maximize IoU during assignment (previously could choose the worst matches), which can affect TP/FP classification and mAP when use_scipy=True.Results.save() 💾📁
create_dir parameter (default True) to control auto-creation of parent directories when saving annotated outputs.torch>=1.13 on Linux.Results.save(create_dir=True) makes saving predictions into new folders simpler (fewer “directory not found” errors).If you’re deploying/training with mixed photo sources (phones, HEIC libraries, multilingual labels), 8.4.15 is a practical upgrade 🛠️🚀
Results.save() with pathlib and optional directory creation by @ShuaiLYU in https://github.com/ultralytics/ultralytics/pull/23592match_predictions to maximize IoU by @Mr-Neutr0n in https://github.com/ultralytics/ultralytics/pull/23634metadata.distribution by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/23641build_reference.py by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/23666CustomSaveTrainer example in custom trainer guide by @raimbekovm in https://github.com/ultralytics/ultralytics/pull/23668ultralytics 8.4.15 HEIC/HEIF image support by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/23714Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.4.14...v8.4.15
Ultralytics 8.4.14 adds Ultralytics Platform “Cancel training” support so you can stop runs quickly *and still keep/upload partial results* ⛔️📤
Ultralytics 8.4.14 adds Ultralytics Platform “Cancel training” support so you can stop runs quickly and still keep/upload partial results ⛔️📤
trainer.stop=True when cancelled=true.trainer._platform_cancelled and logs clear messages.self.stop is set, allowing external stops (like Platform cancellation) to take effect sooner.rle_loss to zero to prevent loss from going negative and destabilizing training (helps avoid mAP dropping across epochs).segment2box() now excludes points exactly on the image border after clipping, avoiding boxes incorrectly snapping to edges (better box regression + mask quality).TypeError when fitness is present but None during tuning (safer/cleaner tuning runs).model.fuse() 📚
If you train via the Ultralytics Platform, this release is especially impactful due to the new cancellation behavior 🛑.
segment2box boundary condition for edge-snapped segment points by @Mr-Neutr0n in https://github.com/ultralytics/ultralytics/pull/23602model.fuse() by @raimbekovm in https://github.com/ultralytics/ultralytics/pull/23601ultralytics 8.4.14 Platform training cancel feature by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/23614Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.4.13...v8.4.14
One column per quarter.