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 7 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11 versions withdrawn
withdrawn after publishing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.
CoreML export on iOS gets more reliable for detection models, fixing NMS issues with non-80-class datasets and simplifying model export handling. Plus
CoreML export on iOS gets more reliable for detection models, fixing NMS issues with non-80-class datasets and simplifying model export handling. Plus, clearer download error messages for easier troubleshooting. 🚀
mlprogram path for iOS detection exports to handle MLProgram-specific behavior.ct_model.weights_dir directly—no manual save-path juggling.safe_download, including underlying exception details for faster debugging.Minimal example (CoreML export with NMS):
yolo export model=yolo11n.pt format=coreml nms=True
# Now more reliable on iOS (MLProgram), even for non-80-class datasets 🎯
ultralytics 8.3.207 Apple MLProgram NMS models 80-class pad workaround by @Y-T-G in https://github.com/ultralytics/ultralytics/pull/22310Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.3.206...v8.3.207
Your coding agent can read these notes before it upgrades. Set up the MCP server →
CoreML export gets a big upgrade: dynamic image shapes and multi-image batching now work end-to-end, making Apple deployments more flexible and produc
CoreML export gets a big upgrade: dynamic image shapes and multi-image batching now work end-to-end, making Apple deployments more flexible and production-friendly. 🚀
dynamic=True for variable image sizes and batch sizes. 🧩dynamic with nms, and classification models can’t use dynamic.dynamic for CoreML.YOLO, YOLOWorld, YOLOE, SAM, etc.) via TYPE_CHECKING for better IDE autocomplete and static analysis. ✨ (@Y-T-G)plot_images now gracefully handles 1-, 2-, 3-, and multi-channel inputs (auto-pads or crops to 3 channels), improving reliability for multispectral and non-RGB data. 🖼️ (@ambitious-octopus)dynamic + nms) and save debugging time.Quick start examples:
# Export CoreML with dynamic shapes and set a max batch
yolo export format=coreml dynamic batch=16
# Batch inference with a CoreML package
yolo predict model=yolo11n.mlpackage batch=2
# Change image size on the fly (still batched)
yolo predict model=yolo11n.mlpackage imgsz=256 batch=2
Notes:
dynamic=True for batch > 1.dynamic=True with nms=True.dynamic=True on CoreML.python.md Docs by @cdeil in https://github.com/ultralytics/ultralytics/pull/22285GIF to AVIF conversion, smaller files size and same animation by @RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/22293plot_images by @ambitious-octopus in https://github.com/ultralytics/ultralytics/pull/22282ignore_init_summary in mkdocs.yml by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/22305ultralytics 8.3.206 Support dynamic batch and image size with CoreML export by @Y-T-G in https://github.com/ultralytics/ultralytics/pull/22300Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.3.205...v8.3.206
Cleaner post-training behavior and clearer visuals: v8.3.205 refines how training configs are restored from checkpoints, improves fitness plots with s
Cleaner post-training behavior and clearer visuals: v8.3.205 refines how training configs are restored from checkpoints, improves fitness plots with smarter outlier filtering, and updates docs for more predictable inference and easy-start notebooks. ✅
_reset_ckpt_args) instead of using raw model.args.Useful links:
predict.rect behavior note to predict.md by @Y-T-G in https://github.com/ultralytics/ultralytics/pull/22266Construction-PPE notebook in docs by @RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/22269ultralytics 8.3.205 Reset checkpoint arguments after training by @Y-T-G in https://github.com/ultralytics/ultralytics/pull/22286Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.3.204...v8.3.205
YOLO 8.3.204 sharpens training ergonomics, stabilizes exports, broadens device support, and refreshes docs—making multi-GPU projects, ONNX workflows,
YOLO 8.3.204 sharpens training ergonomics, stabilizes exports, broadens device support, and refreshes docs—making multi-GPU projects, ONNX workflows, and edge deployments smoother than ever. 🚀
select_device() and into the Trainer, requiring explicit batch sizes when using more than one GPU and delivering clearer error guidance.YOLOESegTrainerFromScratch, cleaner loss initialization, and shared preprocessing streamline from-scratch and fine-tune training paths.non_blocking transfers on MPS.strip_optimizer() now removes AMP scaler state, trimming checkpoint clutter.imgsz, reducing mismatches in production.AttributeError error for validation with visualize by @Y-T-G in https://github.com/ultralytics/ultralytics/pull/22194--slow flag in ci.yml pip install command by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/22196uv version 0.8.19 restraint by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/22197non_blocking device transfers for MPS to prevent output corruption by @Y-T-G in https://github.com/ultralytics/ultralytics/pull/22181torch.compile by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/22205preprocess_batch by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/22206RKNN model support in streamlit live-inference solution by @RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/22212imgsz for visual prompt extraction by @Y-T-G in https://github.com/ultralytics/ultralytics/pull/22238RKNN model support in streamlit live-inference solution by @lakshanthad in https://github.com/ultralytics/ultralytics/pull/22230ultralytics 8.3.204 Scope batch_size check from select_device() to Trainer by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/22254Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.3.203...v8.3.204
Safer, smoother training resumes with AMP plus sturdier exports and better version checks. This release improves mixed‑precision resume stability, fut
Safer, smoother training resumes with AMP plus sturdier exports and better version checks. This release improves mixed‑precision resume stability, future‑proofs ONNX opset selection, tightens export requirements (OpenVINO/CoreML), and polishes compatibility across models and PyTorch versions. 🚀🛡️
best_onnx_opset and multiple-install checks; improved slug generation for content tabs; corrected CI badge; enhanced Model YAML Guide and nav PRs #22142, #22157, #22184, #22154 📚Tip: If you export to OpenVINO or CoreML, verify your PyTorch version first:
import torch
print(torch.__version__)
Upgrade instructions: see the official PyTorch install guide on the PyTorch website.
best_save_dir by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/22137compile for WorldModel and YOLOEModel by @Y-T-G in https://github.com/ultralytics/ultralytics/pull/22148ultralytics 8.3.203 Future-proof ONNX opset selection by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/22142torch<1.13 by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/22155torch>=2.1 for OpenVINO exports by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/22156torch==1.10 by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/22158torch>=1.13 for YOLOE models by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/22160torch>=1.11 for RTDETR models by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/22161torch<1.11 by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/22163torch>=1.11 for CoreML exports by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/22162sum(0).sum() predictor fix by @Y-T-G in https://github.com/ultralytics/ultralytics/pull/22169uv required-version = "==0.8.19" by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/22186ultralytics 8.3.203 Restore GradScaler on resuming training by @Y-T-G in https://github.com/ultralytics/ultralytics/pull/22189Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.3.202...v8.3.203
TFLite export gets a targeted INT8 upgrade: per-channel quantization is fixed and INT8 exports are slimmer and more reliable in ultralytics 8.3.202. 🚀
TFLite export gets a targeted INT8 upgrade: per-channel quantization is fixed and INT8 exports are slimmer and more reliable in ultralytics 8.3.202. 🚀
Quick tip to export INT8 TFLite with the new behavior:
from ultralytics import YOLO
model = YOLO("yolo11n.pt") # or your trained YOLO11 model
model.export(format="tflite", int8=True) # per-channel INT8 fix applied; leaner artifacts
Enjoy improved TFLite quantization and leaner edge-ready models! ✨
ultralytics 8.3.202 TFLite per-channel INT8 quantization fix by @Y-T-G in https://github.com/ultralytics/ultralytics/pull/22133Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.3.201...v8.3.202
Ultralytics 8.3.201 focuses on smoother TensorFlow exports, more reliable Docker-based exports, improved CLI help for Ultralytics HUB, and helpful doc
Ultralytics 8.3.201 focuses on smoother TensorFlow exports, more reliable Docker-based exports, improved CLI help for Ultralytics HUB, and helpful docs refinements — all aimed at making training and deployment more stable and user-friendly. 🚀
TensorFlow export compatibility expanded (primary change) 🔧
ai-edge-litert requirement to >=1.2.0 for non-macOS platforms; macOS remains pinned to <1.4.0 to avoid known issues.check_requirements() now accepts tuple[str] as well as Path | str | list[str].Export robustness and Docker reliability 🐳
tmp/yolo11n.pt) for EdgeTPU/NCNN/IMX steps, improving stability and reproducibility: Export in Dockerfile /tmp dir by @glenn-jocher.Better CLI help for Ultralytics HUB 🧭
yolo -h now shows the correct context-aware help for HUB subcommands by adjusting help precedence: Display help message with yolo -h by @Y-T-G.Training flexibility for multimodal (YOLOE) workflows 🧠
get_text_pe() during training: Remove training mode assertion in get_text_pe() by @Y-T-G.Dataset reliability fix 📦
Documentation improvements 📚
Smoother TensorFlow export workflows ✅
ai-edge-litert. 🛡️More reliable and portable Docker builds 🔁
Clearer, more helpful CLI and logs 🗺️
yolo -h behaves as users expect, especially for Ultralytics HUB commands.Enhanced training flexibility and dataset setup 🧪
Easier customization and configuration 📘
Tip: To try TensorFlow SavedModel export with the updated dependency handling:
pip install ultralytics
yolo export model=yolo11n.pt format=saved_model
get_text_pe() by @Y-T-G in https://github.com/ultralytics/ultralytics/pull/22076/tmp dir by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/22107yolo -h by @Y-T-G in https://github.com/ultralytics/ultralytics/pull/22102ultralytics 8.3.201 Expand TensorFlow ai-edge-litert≥1.2 support (non-macOS) by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/22110Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.3.200...v8.3.201
A focused export refactor that adds a complete Sony IMX500 pipeline, cleans up IMX-specific logic in model heads, and introduces clearer export APIs—p
A focused export refactor that adds a complete Sony IMX500 pipeline, cleans up IMX-specific logic in model heads, and introduces clearer export APIs—plus stability fixes for ONNX (OBB), TensorRT exports, and smoother multi-GPU training. 🚀
Export refactor and new IMX pipeline (PR #22005) ✨
export_onnx → torch2onnxexport_engine → onnx2enginetorch2imx utility with quantization and IMX-friendly NMS wrappingFXModel, Detect/Pose overrides, NMSWrapper)ultralytics/utils/export/ with imx.pyFXModel from torch_utilsIMX support clarity (PR #22082) 🧭
ONNX + OBB reliability (PR #22079) 🧩
TensorRT export stability (PR #22080) ⚙️
DDP training improvements (PR #22073) 💼
world_size/ddp earlier, simplifies DDP APIs, and prevents duplicate wandb logsTuner robustness (PR #22066) 🛡️
KeyError when close_mosaic isn’t in the hyperparameter search spaceUX polish (PRs #22071, #22083, #22046, #22063, #22085) ✍️
pttorch2onnx and onnx2engine names make conversion stages explicit and reduce confusion.torch2imx enables end-to-end IMX500 deployment with built-in quantization and NMS wrapping.Quick tips:
yolo export model=yolo11n.pt format=onnxyolo export model=yolo11n.pt format=engine dynamicyolo export model=yolo11n.pt format=imxKeyError when close_mosaic not part of search space by @Y-T-G in https://github.com/ultralytics/ultralytics/pull/22066pt format by @RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/22083IMX export assert message to include YOLO11n by @RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/22082RuntimeError for OBB models exported with NMS by @Y-T-G in https://github.com/ultralytics/ultralytics/pull/22079wandb logging twice when running DDP training by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/22073ultralytics 8.3.200 Refactor Sony IMX exports to imx.py by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/22005Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.3.199...v8.3.200
Replaces deprecated NVIDIA Docker approach with NVIDIA Container Toolkit; adds distro-specific install steps and standardizes --runtime=nvidia. See PR…
ultralytics 8.3.199 boosts startup speed with lazy model loading, refines export/runtime stability, and modernizes GPU Docker docs—delivering faster imports, smoother deployments, and clearer tooling. ⚡🐳
__getattr__, preserving the public API. About ~3% faster import ultralytics.(boxes, scores, labels, n_valid) for non-keypoint tasks; keypoint outputs unchanged.tensorrt-cu12) and blocks known-bad versions for more reliable exports.attempt_compile() now warns on mode="max-autotune" and uses max-autotune-no-cudagraphs instead; docs updated accordingly.plot_tune_results(..., exclude_zero_fitness_points=True) for cleaner visuals.scale errors in parse_model() when scales isn’t provided.--runtime=nvidia.gpu-latest; Slack alerts now target specific failed jobs; runner image version parameterized.ultralytics/__init__.py.from ultralytics import YOLOultralytics.YOLO("yolo11n.pt")torch.compile now prefers max-autotune-no-cudagraphs, avoiding CUDA Graphs issues while keeping performance benefits.--runtime=nvidia examples make GPU containers more predictable across distros.Helpful snippets:
from ultralytics import YOLO
model = YOLO("yolo11n.pt")
from ultralytics.utils.plotting import plot_tune_results
plot_tune_results("tune_results.csv", exclude_zero_fitness_points=False)
sudo docker run -it --ipc=host --runtime=nvidia --gpus all ultralytics/ultralytics:latest
imx object detection export outputs by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/22045mode="max-autotune" with compile by @Y-T-G in https://github.com/ultralytics/ultralytics/pull/22040parse_model() by @Y-T-G in https://github.com/ultralytics/ultralytics/pull/22054ultralytics 8.3.199 3% Faster Ultralytics Imports with Lazy Model Loading by @RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/21985Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.3.198...v8.3.199
Smarter hyperparameter tuning and sturdier training/inference across the board. v8.3.198 introduces a much stronger Tuner with BLX-α crossover, unifie
Smarter hyperparameter tuning and sturdier training/inference across the board. v8.3.198 introduces a much stronger Tuner with BLX-α crossover, unified metric plotting, safer defaults, and multiple robustness fixes (NMS, DDP loss, Intel GPU checks), plus simpler export APIs and flexible torch.compile modes. 🚀
Quick examples
from ultralytics import YOLO
model = YOLO("yolo11s.yaml")
model.tune(
device=0,
data="coco128.yaml",
optimizer="AdamW",
epochs=100,
batch=8,
compile=False,
plots=False,
val=False,
save=False,
workers=16,
project="tune-yolo11s-scratch-coco128-100e",
iterations=1000,
)
from ultralytics import YOLO
model = YOLO("yolo11n.pt")
model.train(data="coco8.yaml", epochs=3, compile="reduce-overhead") # or "default", "max-autotune", True/False
from ultralytics import YOLO
model = YOLO("yolo11n.pt")
onnx_path = model.export(format="onnx") # 'onnx_path' is a string
Exporter and remove unnecessary None placeholder by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/22009torch.compile by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/21999TorchNMS.nms() early exit logic by @Y-T-G in https://github.com/ultralytics/ultralytics/pull/22014numpy version by @Y-T-G in https://github.com/ultralytics/ultralytics/pull/22025torch.Tensor.shape[0] to get length for torch.Tensor by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/22021plot_results for detect/segment/pose/obb/classify tasks by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/22026mask_ratio=1 by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/22037ultralytics 8.3.198 Improve Tuner with BLX-α gene crossover by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/22038Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.3.197...v8.3.198
Ultralytics 8.3.197 adds the new Construction-PPE dataset and delivers training, export, plotting, HUB, and docs improvements that make YOLO11 workflo
Ultralytics 8.3.197 adds the new Construction-PPE dataset and delivers training, export, plotting, HUB, and docs improvements that make YOLO11 workflows more robust and easier to use. 🦺🚀
preds = model(batch["img"]); loss = model.loss(batch, preds)), removed compile-disabling wrappers, and cleaned up dynamic tensor handling for detect/pose/segment. ⚙️exit_early switch to Fast-NMS and set exit_early=False during export to prevent tracing breaks and empty outputs in OBB and some INT8 backends. 📦max_det instead of a fixed 50-mask cap, giving you consistent, configurable visualizations. 🖼️Helpful links:
max_det settings for more intuitive validation; tune max_det if plotting slows. 🎛️Enjoy the new dataset and smoother YOLO11 experience! 🎉
max_det by @Y-T-G in https://github.com/ultralytics/ultralytics/pull/21992disable_dynamo wrapper around Detect.inference() and loss calculation by @Y-T-G in https://github.com/ultralytics/ultralytics/pull/21996ultralytics 8.3.197 New Construction-PPE dataset by @RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/21976Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.3.196...v8.3.197
Minor typo fix in a deprecation warning.
Optional torch.compile acceleration lands across train/val/predict for up to ~30% faster runs, plus dataloader throughput boosts, CoreML export reliability, unified device handling, and smoother setup/CI/paths. ⚡️🧩
torch.compile acceleration (primary)
compile arg in train/val/predict (default False) with end-to-end wiring via config/CLI/API.attempt_compile(...) to safely enable compile and disable_dynamo(...) to opt out specific code paths.compile=True for accelerated inference.de_parallel ➝ unwrap_model (handles both parallel and compiled models).Faster data loading
prefetch_factor to 4 when num_workers > 0 and seamlessly removed it for older PyTorch (<2.0) to avoid errors.drop_last behavior during compile-enabled training to improve shape stability.Unified, safer device handling
CoreML export robustness
Plotting stability
@plt_settings() to feature_visualization(...) for backend-safe, non-blocking feature map plots.Config directory resolution
get_user_config_dir(): honors YOLO_CONFIG_DIR, follows OS conventions (XDG on Linux), and gracefully falls back to writable paths like /tmp.Compatibility and CI polish
torch.compile; inference can also benefit on supported devices (CUDA/CPU/MPS). 🚀yolo train model=yolo11n.pt data=coco8.yaml epochs=100 compile=Truefrom ultralytics import YOLO
model = YOLO("yolo11n.pt")
model.train(data="coco8.yaml", epochs=100, compile=True)
# Also works for val/predict:
model.val(compile=True)
model.predict("img.jpg", compile=True)
Primary PR: “ultralytics 8.3.196 torch.compile acceleration” by @glenn-jocher (adds compile flag, utilities, and engine integrations).
@plt_settings() decorator to feature_visualization() by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/21973prefetch_factor to 4 by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/21974preprocess_batch methods by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/21981deprecation_warn method by @RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/21987ultralytics 8.3.196 torch.compile acceleration for 30% faster training by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/21975Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.3.195...v8.3.196
Ultralytics 8.3.195 boosts startup speed and reliability by replacing the py-cpuinfo dependency with a built-in, cross‑platform CPU info utility—plus
Ultralytics 8.3.195 boosts startup speed and reliability by replacing the py-cpuinfo dependency with a built-in, cross‑platform CPU info utility—plus several quality fixes for checkpoints, OBB predictions, and dataset verification. 🚀
Zero-dependency CPU info (priority)
ultralytics.utils.cpu.CPUInfo for fast, consistent CPU detection across macOS/Linux/Windows. See reference docs for CPUInfo added to the site.get_cpu_info() to use CPUInfo.name().py-cpuinfo from project dependencies.Checkpoint loading improved
model.args now preserves the full args dict; DEFAULT_CFG_KEYS filter removed).OBB prediction fix
xywh=True in scale_boxes to prevent width/height distortion.Dataset verification bugfix
verify_image_label using the correct keypoint parameter (prevents shape errors on pose datasets).keypoints typo by @LH-and-FPGA (view PR #21965) 🩹Typing modernization
typing generics (List/Dict/...) to built-in list/dict/... across code and examples (no behavior change).Faster startup and leaner installs
More reliable training/resume/export
model.args may contain additional keys—ensure downstream code can handle them gracefully ⚠️More accurate predictions and verification
xywh boxes prevents subtle geometry errors in outputs ✅Quick tip:
from ultralytics.utils.cpu import CPUInfo
print(CPUInfo.name()) # e.g., "Intel(R) Core(TM) i7-8750H CPU @ 2.20GHz"
Enjoy a snappier, more robust Ultralytics experience with 8.3.195! ✨
keypoints typo in verify_image_label by @LH-and-FPGA in https://github.com/ultralytics/ultralytics/pull/21965ultralytics 8.3.195 Zero-dependency 30X faster CPU info by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/21972Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.3.194...v8.3.195
Enjoy improved stability and smoother workflows across exports, CoreML inference, and telemetry—with minimal breaking changes. 🎉
Ultralytics 8.3.194 enhances model export stability and visibility, adds seamless CoreML NMS inference, and refactors telemetry for a cleaner, non-blocking experience—plus several quality-of-life fixes and docs updates. 🚀
onnxslim>=0.1.67 for ONNX and TensorFlow SavedModel exports and restored useful urllib3 logs while keeping sentry_sdk quiet. See PR ultralytics 8.3.194: bump onnxslim in Exporter. 🔧tensorflow>=2.0.0,<=2.19.0 and tf_keras<=2.19.0 to match onnx2tf requirements. See PR Pin tf_keras<=2.19.0. 🧩ultralytics.utils.events module; sending runs in a background thread and no longer depends on HUB. See PR Refactor Events class. 🧠cv2.resize channel limit (512) by splitting large mask tensors to prevent errors with many detections/classes. See PR Handle cv2.resize 512-channel limit. 🛡️source and renamed img param to im in inference, updating examples accordingly. See PR Fix SAM2DynamicInteractivePredictor example. 📘onnxslim brings bug fixes and improved model simplification for both ONNX and TensorFlow exports. Restored urllib3 logs help troubleshoot network issues during export. ✅yolo export model=yolo11n.pt format=coreml nms=True
yolo predict model=yolo11n.mlpackage
ultralytics.utils.events and sent in the background—no HUB dependency and zero impact on training/export performance. If you import Events directly, update your import:from ultralytics.utils.events import events, Events
source=... consistently, reducing confusion. 🧭Enjoy improved stability and smoother workflows across exports, CoreML inference, and telemetry—with minimal breaking changes. 🎉
tf_keras<=2.19.0 by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/21949cv2.resize 512 limitations by @ShuaiLYU in https://github.com/ultralytics/ultralytics/pull/21947SAM2DynamicInteractivePredictor example in docs by @Y-T-G in https://github.com/ultralytics/ultralytics/pull/21955ultralytics 8.3.194 Bump onnxslim>=0.1.67 in Exporter by @onuralpszr in https://github.com/ultralytics/ultralytics/pull/21951Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.3.193...v8.3.194
Ultralytics 8.3.193 boosts long-video and large-batch inference performance by auto-enabling TorchVision NMS, streamlines checkpoint loading with a si
Ultralytics 8.3.193 boosts long-video and large-batch inference performance by auto-enabling TorchVision NMS, streamlines checkpoint loading with a single API, and delivers cleaner, more informative progress bars and logging. 🚀
torchvision to trigger TorchVision-based NMS automatically.attempt_load_one_weight/attempt_load_weights with a single load_checkpoint function across the codebase.ultralytics.nn.tasks.load_checkpoint.from ultralytics.nn.tasks import load_checkpoint
model, ckpt = load_checkpoint("yolo11n.pt", device="cpu", fuse=False)
get_save_dir() now returns absolute paths for unambiguous run locations in logs.onnxslim==0.1.65 for ONNX and TF SavedModel exports to avoid unexpected breakages.TORCH_VERSION usage to prevent issues with non-string version objects and align all version checks/logging.overlap_mask=True and simplifies IoU computation for more accurate, efficient val results.autosplit import in xView.yaml (ultralytics.data.split.autosplit).load_checkpoint entry point; fewer APIs to remember. 🧠Tip: If any tooling relies on exact progress bar strings, review and adjust parsers due to the new TQDM formatting. ✍️
save_dir by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/21931onnxslim==0.1.65 by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/21941autosplit import in xView.yaml by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/21939torch._version_ check for Torch>1.9 by @RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/21921_prepare_batch by @Y-T-G in https://github.com/ultralytics/ultralytics/pull/21503ultralytics 8.3.193 Preload torchvision for long-sequence inference by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/21926Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.3.192...v8.3.193
Distributed hyperparameter tuning lands in ultralytics 8.3.192 with optional MongoDB Atlas coordination, delivering scalable HPO across machines, a st
Distributed hyperparameter tuning lands in ultralytics 8.3.192 with optional MongoDB Atlas coordination, delivering scalable HPO across machines, a steadier progress ETA, faster GPU data transfers, and more reliable CSV parsing — plus a clear docs warning on YOLOv5 compatibility. 🚀
mongodb_uri, mongodb_db (default: ultralytics), mongodb_collection (default: tuner_results)..to(device, non_blocking=True) across train/val for Classification, Detection, Pose, Segmentation, YOLO-World (text embeddings), and YOLOE (text/visual prompts) to reduce data-transfer bottlenecks. ⚡infer_schema_length=None) across dataset conversion, training results, and plotting, preventing type mix-ups and improving plot/metric accuracy. 📈Quick start for distributed tuning:
from ultralytics import YOLO
model = YOLO("yolo11n.pt")
model.tune(
data="coco8.yaml",
epochs=10,
iterations=300,
mongodb_uri="mongodb+srv://user:pass@cluster.mongodb.net/",
mongodb_db="ultralytics",
mongodb_collection="tuner_results",
plots=False,
save=False,
val=False,
)
Helpful links:
non_blocking=True for additional train/val loaders by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/21912ultralytics 8.3.192 Distributed Hyperparameter Tuning with MongoDB Atlas Integration by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/21882Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.3.191...v8.3.192
v8.3.191 makes Git metadata access blazing fast and more reliable, simplifies model-loading internals, and polishes docs, types, and tests—resulting i
v8.3.191 makes Git metadata access blazing fast and more reliable, simplifies model-loading internals, and polishes docs, types, and tests—resulting in a snappier, cleaner developer experience. ⚡🧹
⚡ Pure-pathlib Git metadata (no subprocess)
GitRepo utility in ultralytics/utils/git.py to read branch, commit, and origin directly from the filesystem (incl. worktrees and packed-refs).🧠 AutoBackend/model-loading refinements
attempt_load_one_weight for cleaner single-weight loading and clearer APIs.list[str] weights in AutoBackend; pass a single path or a torch.nn.Module.🧠 Modern types + style cleanup
|, postponed annotations).📘 Documentation improvements
__init__ summaries.autosplit import path in guides (ultralytics.data.split.autosplit).ultralytics.utils.git.🔖 Version bump
__version__ = "8.3.191"Minimal example for the new Git utility:
from ultralytics.utils.git import GitRepo
g = GitRepo() # auto-resolves repo root
if g.is_repo:
print("Branch:", g.branch)
print("Commit:", g.commit)
print("Origin:", g.origin)
list[str] weights to AutoBackend, switch to one model path or load a torch.nn.Module.ignore_init_summary by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/21889ultralytics 8.3.191 Pure-pathlib Git metadata (no subprocess, ~200× faster) by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/21885Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.3.190...v8.3.191
ultralytics 8.3.190 delivers a faster, torchvision-free NMS pipeline powered by a new PyTorch-native implementation optimized for YOLO11, plus stabili
ultralytics 8.3.190 delivers a faster, torchvision-free NMS pipeline powered by a new PyTorch-native implementation optimized for YOLO11, plus stability and UX upgrades across CoreML, progress bars, macOS handling, and reproducibility. 🚀
Autobackend torch-native NMS (most important)
ultralytics/utils/nms.py with:
TorchNMS.nms (exact-match, optimized NMS) and TorchNMS.fast_nms (vectorized FastNMS)non_max_suppression refactored here; rotated NMS supported via fast_nms + batch_probioutorchvision.ops.nms for prediction-time usepredict, val, examples) to from ultralytics.utils import nmsTorchNMS.fast_nms with batch_probiou for rotated boxesAutoBackend API cleanup
weights= to model= for consistency and to avoid duplicate memory usage (@glenn-jocher)torch.nn.Module and .pt loading; gradients disabled by default for safer, faster inferenceCoreML stability
coremltools>=8.0 for CoreML inference to prevent runtime issues; shows a helpful message before loadingPerformance and stability
sigmoid_(), clamp_()) for non-macOS 14 systems to boost speedMACOS14/NOT_MACOS14 flags to handle Apple MPS quirks safelyImport-time and CI improvements
Geometric ops robustness
Reproducibility
git_commit in checkpoints; new utils.get_git_commit() with cachingUX and docs
subprocess.run(...) for safer command executionFaster, leaner inference 🏃♂️💨
torchvision, speeds up post-processing, and lowers latency, especially helpful in deployment and environments where torchvision is heavy or unavailable.Better platform reliability 🛡️
Improved developer experience 👩💻
AutoBackend(model=...) API, unified PyTorch paths, and no-grad defaults reduce boilerplate and risks.Stronger reproducibility and CI coverage 📦
Faster startup and “just works” behavior ⚙️
Quick usage examples:
from ultralytics.utils import nms
# preds: [B, C+4(+M), N]
out = nms.non_max_suppression(preds, conf_thres=0.25, iou_thres=0.45)
from ultralytics.nn.autobackend import AutoBackend
model = AutoBackend(model="yolo11n.pt", device="cuda")
pip install -U "coremltools>=8.0"
coremltools>=8.0 check to Autobackend by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/21861TQDM progress bar to display remaining time by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/21868obb by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/21871requests imports by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/21876MACOS14 check by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/21883weights to model by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/21881.clamp_() and .sigmoid_() ops by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/21878subprocess calls by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/21887subprocess.run() for improved dataset autodownloads by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/21888ultralytics 8.3.190 Autobackend torch-native NMS by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/21862Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.3.189...v8.3.190
ultralytics 8.3.189 delivers a 3x faster in-place model.fuse() for leaner, quicker inference, plus better NVIDIA Jetson handling with version-aware de
ultralytics 8.3.189 delivers a 3x faster in-place model.fuse() for leaner, quicker inference, plus better NVIDIA Jetson handling with version-aware detection and more robust autobackend behavior. 🚀
In-place layer fusion (primary change) ⚡
Jetson-aware autobackend improvements ✅
Version-aware Jetson detection 🔍
Version bump
Faster inference setups and smaller memory footprint 🚀
More reliable Jetson deployments, especially on JetPack 5 🧩
Easier, precise hardware targeting 🛠️
Backward compatible and safer defaults ✅
Quick examples:
Faster in-place fusion
from ultralytics import YOLO
model = YOLO('yolov8n.pt').model
model = model.fuse() # now runs ~3x faster and in-place
Jetson version checks
from ultralytics.utils import is_jetson
if is_jetson(): # Any Jetson
...
if is_jetson(5): # Specifically JetPack 5
...
Happy building and deploying! ✨
model.fuse() order of operations speedup by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/21853ultralytics 8.3.189 3x faster inplace model.fuse() by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/21844Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.3.188...v8.3.189
Faster, more reliable downloads and smoother developer workflows, plus improved export compatibility and clearer JSON outputs — all wrapped in Ultraly
Faster, more reliable downloads and smoother developer workflows, plus improved export compatibility and clearer JSON outputs — all wrapped in Ultralytics 8.3.188. 🚀
<1.18.0 due to TensorFlow hangs.onnxslim>=0.1.65."file_name" across RT-DETR, YOLO Detect, and YOLO OBB validators.total=0 gracefully; richer type hints for better DX.file_name in predictions.json simplifies mapping outputs back to source images. 🔎Upgrade with:
pip install -U ultralytics
sam-2.md by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/21835file_name field to predictions.json for better image-annotation alignment by @olena-hul-dataspan in https://github.com/ultralytics/ultralytics/pull/21837onnx>=1.18.0 by @Y-T-G in https://github.com/ultralytics/ultralytics/pull/21758Linux by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/21848TQDM progress bars when working with empty sequence by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/21849ultralytics 8.3.188 Faster downloads via a single request by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/21850Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.3.187...v8.3.188
Impact: Quicker data exports and plotting; smaller dependency footprint. Breaking change: code expecting Pandas DataFrames from to_df() must adapt to…
SAM 2 gets a powerful, training‑free upgrade for interactive, few‑shot multi-object segmentation and tracking, while Ultralytics switches core data operations to Polars for faster, lighter analytics. 🚀
SAM2DynamicInteractivePredictor (new, priority)
Data layer refactor: Pandas ➜ Polars
to_df() now returns a Polars DataFrame; CSV/JSON exports use Polars.polars, removed pandas and pandas-stubs.Visualization and UX
plot_tune_results() gains consistent Matplotlib styling via @plt_settings() for headless and notebook runs.Documentation improvements
mkdocs-ultralytics-plugin>=0.1.29.Interactive, few-shot segmentation/tracking (SAM2DynamicInteractivePredictor)
from ultralytics.models.sam import SAM2DynamicInteractivePredictor
overrides = dict(model="sam2_t.pt", task="segment", mode="predict", imgsz=1024, conf=0.01, save=False)
predictor = SAM2DynamicInteractivePredictor(overrides=overrides, max_obj_num=10)
# Add an object with a box prompt and store in memory
predictor.inference(img="image1.jpg", bboxes=[[100, 100, 200, 200]], obj_ids=[1], update_memory=True)
# Track it in a new image
results = predictor(source="image2.jpg")
# Add another object later
predictor.inference(img="image3.jpg", bboxes=[[300, 300, 400, 400]], obj_ids=[2], update_memory=True)
# Continue inference
results = predictor(source="image4.jpg")
Faster, lighter data workflows with Polars
to_df() must adapt to Polars (e.g., df.to_pandas() if needed). Removed XML/HTML/SQL export helpers—use CSV/JSON or Polars I/O instead.Better plots and docs
Useful links:
pandas to polars (training, benchmarks, plotting, W&B) by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/21619mkdocs-ultralytics-plugin for beautifulsoup4>=4.13.5 by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/21811plt_settings decorator to plot_tune_results for agg backend by @onuralpszr in https://github.com/ultralytics/ultralytics/pull/21818mobile-sam.md by @RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/21825instances count on bars in labels.jpg by @darouwan in https://github.com/ultralytics/ultralytics/pull/21784ultralytics 8.3.187 SAM2: Add SAM2DynamicInteractivePredictor support few-shot inference by @ShuaiLYU in https://github.com/ultralytics/ultralytics/pull/21232Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.3.186...v8.3.187
Lean and polished progress bars, sturdier downloads, and smarter GPU handling. v8.3.186 replaces third‑party tqdm with a fast, zero‑dependency TQDM, i
Lean and polished progress bars, sturdier downloads, and smarter GPU handling. v8.3.186 replaces third‑party tqdm with a fast, zero‑dependency TQDM, improves reliability across downloads and exports, and streamlines CI/docs. 🚀
ultralytics.utils.tqdm.TQDM with iterator/context support and clean single‑line, rich‑style output.TQDM.safe_download() detects partial downloads via Content-Length, logs clear warnings, and cleans incomplete files between retries.pin_memory is now enabled only on CUDA systems to avoid unnecessary warnings on CPU-only setups.pynvml to NVIDIA’s official nvidia-ml-py package; CI installs it for GPU tests (PR #21792).!uv pip install ... for speed and reliability.Minimal example with the new progress bar:
from ultralytics.utils import TQDM
for _ in TQDM(range(100), desc="Processing", unit="items"):
...
See the Ultralytics Docs for details: https://docs.ultralytics.com
pynvml on GPU runners by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/21792safe_download for partial downloads by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/21791!uv pip install in jupyter notebooks by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/21793safe_downloads by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/21794click==8.2.1 from docs.yml by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/21797pynvml with NVIDIA official nvidia-ml-py package by @onuralpszr in https://github.com/ultralytics/ultralytics/pull/21795region visualization in zone-counting solution by @RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/21805coreml export device bug on macOS and linux by @RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/21802ultralytics 8.3.186 Zero-dependency TQDM progress bars by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/21790Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.3.185...v8.3.186
Improved training progress bars for Rich TQDM users, plus robustness and CI/documentation polish for a smoother overall experience. 🎛️✨
Improved training progress bars for Rich TQDM users, plus robustness and CI/documentation polish for a smoother overall experience. 🎛️✨
YOLO_TQDM_RICH=true. ✅YOLODataset now defaults to 3 image channels (RGB) if not provided.yolo_bbox2segment() explicitly sets channels=3 to avoid missing-key errors.uv pip install and uv pip list.pynvml, tensorrt, onnxruntime-gpu) to streamline runs.SystemLogger.get_metrics() docstring with accurate Python examples.YOLO_TQDM_RICH=true is set.Enable Rich progress bars:
YOLO_TQDM_RICH=true yolo train ...import os
os.environ["YOLO_TQDM_RICH"] = "true"
from ultralytics import YOLO
YOLO("yolo11n.pt").train(data="coco8.yaml", epochs=1)
See the Ultralytics Docs for details: https://docs.ultralytics.com 🚀
uv pip install GPU CI requirements by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/21776yolo_bbox2segment by @Y-T-G in https://github.com/ultralytics/ultralytics/pull/21772neuralmagic blog redirect from integration page by @RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/21787Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.3.184...v8.3.185
v8.3.184 introduces a new SystemLogger for real-time system metrics and a high-performance ConsoleLogger for early, continuous training log capture —
v8.3.184 introduces a new SystemLogger for real-time system metrics and a high-performance ConsoleLogger for early, continuous training log capture — improving observability, reliability, and integrations across YOLO11 workflows. 📈🖥️📝
Quick start examples:
from ultralytics.utils.logger import ConsoleLogger
logger = ConsoleLogger("train.log")
logger.start_capture()
print("Training starts...")
logger.stop_capture()
from ultralytics.utils.logger import SystemLogger
syslog = SystemLogger()
print(syslog.get_metrics()) # {'cpu': ..., 'ram': ..., 'disk': {...}, 'network': {...}, 'gpus': {...}}
See the Ultralytics Docs for details.
async functions in Docs bug by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/21760ultralytics 8.3.184 New SystemLogger class by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/21764Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.3.183...v8.3.184
Train YOLO11 models directly from NDJSON datasets with one command — including automatic conversion and image downloading — plus improved YOLOE docs/e
Train YOLO11 models directly from NDJSON datasets with one command — including automatic conversion and image downloading — plus improved YOLOE docs/examples and more reliable GPU CI on GitHub. 🚀
convert_ndjson_to_yolo(ndjson_path, output_path=None) by @glenn-jocher.ndjson file to model.train(data=...) and it auto-converts and generates data.yaml ✅single_cls=True, plus quieter device selection logs by @ShuaiLYU 📚Minimal examples:
from ultralytics import YOLO
model = YOLO("yolo11n.pt")
model.train(data="path/to/dataset.ndjson", epochs=100, imgsz=640)
yolo detect train data=path/to/dataset.ndjson model=yolo11n.pt epochs=100 imgsz=640
See the Ultralytics Docs for details.
ultralytics 8.3.183 New NDJSON dataset format support by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/21747Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.3.182...v8.3.183
Ultralytics 8.3.182 declares the package as “partially typed” for better IDE/type-checker support, plus key robustness and usability tweaks for SAM im
Ultralytics 8.3.182 declares the package as “partially typed” for better IDE/type-checker support, plus key robustness and usability tweaks for SAM image sizing and mask plotting. 🧩✨
Tip: You can now plot masks on CPU without providing a torch image tensor:
See the Ultralytics Docs for details: https://docs.ultralytics.com
imgsz and update rect description in docs by @Y-T-G in https://github.com/ultralytics/ultralytics/pull/21737imgsz for SAM2 tracking by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/21743numpy masks support to Annotator.masks method by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/21744ultralytics 8.3.182 Enable type-checking with py.typed marker by @Y-T-G in https://github.com/ultralytics/ultralytics/pull/21698Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.3.181...v8.3.182
v8.3.181 focuses on safer, faster mixed-precision (FP16) support for SAM/SAM2 and YOLOE pipelines, plus reliability fixes across validation exports, s
v8.3.181 focuses on safer, faster mixed-precision (FP16) support for SAM/SAM2 and YOLOE pipelines, plus reliability fixes across validation exports, source loading, and single-class training. 🚀
SAM FP16 support and dtype/device consistency (PR #21735, priority)
SAM2 robustness without high_res_features (PR #21726)
YOLOE device/half propagation and stability (PR #21670)
Validation export consistency across tasks (PR #21719)
YOLOE visual prompt predictor switching fix (PR #21731)
Single-class training compatibility (PR #21725)
CSV source support for inference (PR #21729)
Streamlit Live Inference improvements (PR #21553)
YOLOE docs enhancements (PR #21728)
save_txt coordinates scaling by @Y-T-G in https://github.com/ultralytics/ultralytics/pull/21719Live Inference solution by @RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/21553classes compatibility with single_cls by @Y-T-G in https://github.com/ultralytics/ultralytics/pull/21725predictor not switching after initialization by @Y-T-G in https://github.com/ultralytics/ultralytics/pull/21731device selection and support half inference for visual prompt by @RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/21670high_res_features by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/21726ultralytics 8.3.181 Support half inference for SAM models by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/21735Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.3.180...v8.3.181
SAM/SAM2 get a new feature-based inference path with standardized tensor outputs, smarter prompt scaling, and a cleaner API — making segmentation fast
SAM/SAM2 get a new feature-based inference path with standardized tensor outputs, smarter prompt scaling, and a cleaner API — making segmentation faster to integrate and more consistent end-to-end. 🚀
Predictor example in docs by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/21700ultralytics 8.3.180 new inference_features method for SAM models by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/21708Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.3.179...v8.3.180
Stability-focused patch: pins TensorFlow for reliable exports, streamlines Docker images (including a new export-ready image), improves YOLOE class ha
Stability-focused patch: pins TensorFlow for reliable exports, streamlines Docker images (including a new export-ready image), improves YOLOE class handling, and tightens CI to reduce flakiness. 🚀
Tip: If you export to TensorFlow, ensure your environment respects tensorflow<=2.19.0 and numpy<2.0.0 to match this release. 📤✨
set_classes and support text prompt CLI usage by @RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/21667click version lock as 8.2.2 was yanked by @onuralpszr in https://github.com/ultralytics/ultralytics/pull/21671yolo checks spacing by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/21676ultralytics 8.3.179 Pin tensorflow<=2.19.0 for exports by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/21680Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.3.178...v8.3.179
Lighter, faster Docker images with a clear split between a minimal Python runtime and a full export toolkit, plus improved build workflows, headless d
Lighter, faster Docker images with a clear split between a minimal Python runtime and a full export toolkit, plus improved build workflows, headless deps, and small API enhancements for smoother installs and preprocessing. 🚀🐳
Enjoy faster, cleaner Docker workflows and more configurable pipelines—whether you’re prototyping with YOLO or exporting to edge backends. 🚀✨
opencv-python-headless in Dockerfiles by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/21647padding_value and interpolation in LetterBox for better compatibility by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/21652ultralytics 8.3.178 new lighter Dockerfile-python images by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/21661Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.3.177...v8.3.178
Stronger supply chain transparency and compliance: SBOMs help with audits, vulnerability tracking, and enterprise procurement. 🔍✅
Automated SPDX SBOMs are now generated and attached to every release, improving security and compliance; plus enhanced TorchScript export flexibility with dynamic shapes and expanded MNN export testing. 🔐📦🧪
sahi-tiled-inference guide by @CoderUni in https://github.com/ultralytics/ultralytics/pull/21616dynamic=True for torchscript exports by @lakshanthad in https://github.com/ultralytics/ultralytics/pull/21573- from parking-management.md FAQs by @RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/21640ultralytics 8.3.177 Add automated SPDX SBOM generation to releases by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/21643Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.3.176...v8.3.177
No expected breaking changes for most users. 🎉
Ultralytics 8.3.176 improves experiment tracking with Comet ML, streamlines YOLOE exports, and refactors tracking internals for cleaner, more flexible pipelines—plus a new CoreML video to speed up Apple deployments. 🚀📊
No expected breaking changes for most users. 🎉
Results object to filter candidates for tracker by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/21536ultralytics 8.3.176 Comet integration fix for YOLO Detect and OBB models by @yaricom in https://github.com/ultralytics/ultralytics/pull/21613Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.3.175...v8.3.176
This release enhances model safety, improves documentation and logging, expands export testing, and updates example notebooks for a smoother and more
This release enhances model safety, improves documentation and logging, expands export testing, and updates example notebooks for a smoother and more reliable user experience. 🚦📚📝
This update is recommended for all users who want safer workflows, better documentation, and a more seamless experience with Ultralytics models and tools. 🚀
classes and total counts in solutions Logging by @RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/21483--slow CI by @lakshanthad in https://github.com/ultralytics/ultralytics/pull/21572queue-management notebook in docs by @RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/21489batch_size for TensorRT inference by @syedhamzamohiuddin in https://github.com/ultralytics/ultralytics/pull/21592ultralytics 8.3.175 YOLOE is_fused() check prior to setting classes by @Y-T-G in https://github.com/ultralytics/ultralytics/pull/21605Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.3.174...v8.3.175
This release focuses on streamlining development workflows, improving reliability, and enhancing documentation across the Ultralytics ecosystem. 🚀🛠️
This release focuses on streamlining development workflows, improving reliability, and enhancing documentation across the Ultralytics ecosystem. 🚀🛠️
uv tool for much faster Python package installation ⚡Overall, this update makes the Ultralytics development process faster, more transparent, and more robust—benefiting both the community and end users. 🎉
ultralytics 8.3.174 Streamlined Conda package CI by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/21577Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.3.173...v8.3.174
This release improves how training metrics are logged to ClearML, making results easier to read and analyze. It also updates documentation tools for a
This release improves how training metrics are logged to ClearML, making results easier to read and analyze. It also updates documentation tools for a smoother developer experience. 🚀📚
mkdocs-ultralytics-plugin to version 0.1.26 for better documentation features.click library to version 8.2.1 to resolve a MkDocs build issue.mkdocs-ultralytics-plugin>=0.1.26 by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/21565click==8.2.1 to fix broken MkDocs build by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/21569ultralytics 8.3.173 Fix ClearML logging for metrics and val categories by @darouwan in https://github.com/ultralytics/ultralytics/pull/21549Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.3.172...v8.3.173
This release streamlines the file download process and updates documentation, making Ultralytics tools lighter and more user-friendly. 🚀
This release streamlines the file download process and updates documentation, making Ultralytics tools lighter and more user-friendly. 🚀
ultralytics 8.3.172 unify downloads progress TQDM by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/21563Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.3.171...v8.3.172
This release brings enhanced Docker workflow security with NVIDIA NGC authentication, a comprehensive new guide for deploying YOLO11 on Google Cloud V
This release brings enhanced Docker workflow security with NVIDIA NGC authentication, a comprehensive new guide for deploying YOLO11 on Google Cloud Vertex AI, improved object tracking consistency, and several documentation and usability updates. 🚀🔒
pydantic dependency for IMX model export to prevent compatibility issues, and clarified export file naming in the documentation. 📦📝Overall, this update strengthens security, usability, and documentation, while making advanced deployment and tracking features more accessible to the community. 🌍✨
metrics.py by @nikunjlad in https://github.com/ultralytics/ultralytics/pull/21526pydantic<=2.11.7 by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/21537ultralytics 8.3.171 add NVIDIA NGC auth to Docker Action by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/21556Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.3.170...v8.3.171
This release streamlines batch processing and inference for Ultralytics models—especially with OpenVINO—while boosting reliability, transparency, and
This release streamlines batch processing and inference for Ultralytics models—especially with OpenVINO—while boosting reliability, transparency, and user experience across the platform. 🚀
Overall, this update makes Ultralytics models easier to use, more robust, and better documented—helping both new and experienced users get the most out of their AI projects! 🌟
NoneType error while reading image by @nikunjlad in https://github.com/ultralytics/ultralytics/pull/21467batch>1 by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/21508ultralytics 8.3.170 Eliminate batch argument for AutoBackend class and fix OpenVINO inference by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/21452Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.3.169...v8.3.170
This release introduces powerful new visualizations for model evaluation, making it easier to understand and debug your results across Detect, Segment
This release introduces powerful new visualizations for model evaluation, making it easier to understand and debug your results across Detect, Segment, Pose, and Oriented Bounding Box (OBB) tasks. 🖼️✨
visualize Option: Added a visualize=True argument for validation commands, enabling these detailed visual breakdowns.Overall, this update brings valuable tools for both beginners and experts to better understand, visualize, and improve their computer vision projects with Ultralytics. 🚀
predictor args while getting model names by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/21443max_det for inference by @atalaydenknalbant in https://github.com/ultralytics/ultralytics/pull/21457optimization_for_gpu_delegate flag for TF exports by @Y-T-G in https://github.com/ultralytics/ultralytics/pull/21482ultralytics 8.3.169 Add GT, TP, FP, FN visualization for Detect, Segment, Pose and OBB by @Y-T-G in https://github.com/ultralytics/ultralytics/pull/18868Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.3.168...v8.3.169
Ultralytics 8.3.168 delivers a major upgrade to how YOLO models export predictions, introduces a unified annotation method, and enhances both document
Ultralytics 8.3.168 delivers a major upgrade to how YOLO models export predictions, introduces a unified annotation method, and enhances both documentation and user-facing tools for a smoother, more consistent experience across the platform. 🚀✨
circle_label and text_label with a single, flexible adaptive_label method for drawing labels (circle or rectangle) on images, simplifying both code and usage.adaptive_label method makes it easier for users and developers to annotate images, with less code and more flexibility.This release is all about making Ultralytics tools more powerful, intuitive, and accessible for everyone—from researchers and developers to everyday users. 💡🖼️🛠️
Ultralytics YOLO11 docs citation section by @RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/21407circle_label and text_label into adaptive_label by @RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/21377text_model module import for Solutions by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/21436image inference support to Streamlit application by @RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/21413ultralytics 8.3.168 Optimize unnecessary native-space calculation by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/21379Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.3.167...v8.3.168
This release enhances Sony IMX device support, improves pose estimation export, and makes visual prompting workflows smoother—while also refining docu
This release enhances Sony IMX device support, improves pose estimation export, and makes visual prompting workflows smoother—while also refining documentation and confusion matrix plotting. 🚀🤖
mct-quantizers>=1.6.0 as a required dependency for IMX model export to ensure compatibility.In summary:
This update makes Ultralytics models more versatile on Sony IMX devices, simplifies visual prompting, and improves the overall user experience for both developers and non-experts. 🎉
self.names overwriting in ConfusionMatrix by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/21356visual_prompts when running prediction with refer_image by @Y-T-G in https://github.com/ultralytics/ultralytics/pull/21368IMX export and inference for Pose Estimation by @ambitious-octopus in https://github.com/ultralytics/ultralytics/pull/20196ultralytics 8.3.167 Fix Sony IMX export mct-quantizers>=1.6.0 dependency by @ambitious-octopus in https://github.com/ultralytics/ultralytics/pull/21404Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.3.166...v8.3.167
This release focuses on improving dataset handling and clarity, with a major update to the VisDrone dataset structure and conversion process, along wi
This release focuses on improving dataset handling and clarity, with a major update to the VisDrone dataset structure and conversion process, along with several refinements to dataset configurations and evaluation logic. 🗂️✨
images/train, images/val, and images/test.Overall, this update streamlines dataset usage, improves evaluation reliability, and enhances the user experience for both development and documentation. 🚀
tiger-pose.yaml download size by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/21355ultralytics 8.3.166 Standardize VisDrone autodownload structure by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/21367Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.3.165...v8.3.166
This release standardizes the folder structure for datasets in Ultralytics, making dataset paths more consistent and user-friendly across multiple YAM
This release standardizes the folder structure for datasets in Ultralytics, making dataset paths more consistent and user-friendly across multiple YAML configuration files. 📁✨
images/train, images/val, and images/test.Overall, this update improves the reliability and usability of dataset management, helping users work more efficiently with Ultralytics models and tools.
ultralytics 8.3.165 Update datasets YAML's by @RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/21353Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.3.164...v8.3.165
This release delivers a critical fix to YOLO detection validation metrics, improves dataset flexibility, enhances export reliability, and polishes doc
This release delivers a critical fix to YOLO detection validation metrics, improves dataset flexibility, enhances export reliability, and polishes documentation and developer experience. 🚀
max_samples parameter for controlling the number of text samples in GroundingDataset and improved logic for negative text selection.valid/ as a fallback validation folder, improving compatibility with Roboflow and similar dataset exports.numpy.ndarray to np.ndarray across code and documentation for clarity.pillow_heif to pi-heif for HEIC image decoding, simplifying license compliance.sweep_annotator method.This update is recommended for all users, especially those validating detection models, working with custom datasets, or exporting models for deployment. 🚀🛠️
max_samples value between GroundingDataset and YOLOMultiModalDataset by @Y-T-G in https://github.com/ultralytics/ultralytics/pull/21286valid/ as fallback validation folder for classification datasets by @JamesBond6873 in https://github.com/ultralytics/ultralytics/pull/21321classify_augmentations by @Toprak2 in https://github.com/ultralytics/ultralytics/pull/21322CSS from similarity-search.html by @RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/21335sweep_annotator example by @RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/21331pillow_heif to pi_heif by @picsalex in https://github.com/ultralytics/ultralytics/pull/21339ultralytics 8.3.164 Fix swapped mAP50 and mAP50-95 in COCOEval stats by @Y-T-G in https://github.com/ultralytics/ultralytics/pull/21350Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.3.163...v8.3.164
This release brings smarter label validation, improved region counting, easier video frame management, and enhanced export and CI workflows for a smoo
This release brings smarter label validation, improved region counting, easier video frame management, and enhanced export and CI workflows for a smoother and more reliable Ultralytics experience. 🚀✨
For All Users:
For Developers & Advanced Users:
Overall, this update makes Ultralytics tools more robust, user-friendly, and ready for diverse real-world applications! 🌍💡
region initialization every frame ~2x faster by @RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/21288save_txt and save_frames by @jugal-sheth in https://github.com/ultralytics/ultralytics/pull/21276onnxslim>=0.1.59 to TOML export dependencies by @inisis in https://github.com/ultralytics/ultralytics/pull/21302onnxslim from pyproject.toml due to Jetson6 docker tests failing by @lakshanthad in https://github.com/ultralytics/ultralytics/pull/21304ultralytics 8.3.163 Add 1% tolerance for labels normalization check by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/21310Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.3.162...v8.3.163
This release (v8.3.162) brings improved reliability and consistency to model loading, enhanced hardware compatibility, and several quality-of-life upd
This release (v8.3.162) brings improved reliability and consistency to model loading, enhanced hardware compatibility, and several quality-of-life updates for both developers and users. 🚀🛠️
torch.load are replaced with Ultralytics' torch_load utility, ensuring consistent and robust model file handling throughout the codebase.ai-edge-litert package is now pinned to versions >=1.2.0,<1.4.0 to ensure stable TensorFlow SavedModel exports.torch_load, users and developers benefit from fewer bugs and more predictable behavior when working with PyTorch models.Overall, v8.3.162 delivers a more robust, user-friendly, and developer-friendly experience across the Ultralytics ecosystem! 🎉
is_lvis check for open-vocabulary models evaluation by @ImJaewooChoi in https://github.com/ultralytics/ultralytics/pull/21245Pose and Segment tasks by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/21259DATASET_DIR relative path compatibility for grounding datasets by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/21256CopyPaste augmentation by @Y-T-G in https://github.com/ultralytics/ultralytics/pull/21262ai-edge-litert>=1.2.0,<1.4.0 by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/21265ultralytics 8.3.162 Replace torch.load calls with patched torch_load method that defaults to weights_only=False by @Y-T-G in https://github.com/ultralytics/ultralytics/pull/21260Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.3.161...v8.3.162
Deprecation Warning for Examples: Added a clear notice that community-contributed examples will be retired in Ultralytics v8.4.0, encouraging users to…
This release focuses on making dataset paths simpler and more consistent across Ultralytics, improving documentation, and enhancing compatibility for various platforms and users. 🗂️✨
coco) instead of legacy relative paths (like ../datasets/coco).datasets_dir locations.This update is all about making Ultralytics easier to use, more robust, and ready for a wider range of users and devices! 🚀
final_mixed_train_no_coco_segm in GroundingDataset by @ImJaewooChoi in https://github.com/ultralytics/ultralytics/pull/21215ultralytics/examples by @Y-T-G in https://github.com/ultralytics/ultralytics/pull/21208visdrone2yolo function by @banu4prasad in https://github.com/ultralytics/ultralytics/pull/21226paddlepaddle==3.0.0 for ARM64 by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/21238summary method in val.md by @RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/21236BaseSolution inheritance in VisualAISearch by @RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/21240ultralytics 8.3.161 Eliminate ../datasets/ in data yaml files for better datasets_dir compatibility by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/21091Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.3.160...v8.3.161
This release brings improved keypoint handling, smarter data augmentation, and enhanced usability for training and exporting models, making YOLO workf
This release brings improved keypoint handling, smarter data augmentation, and enhanced usability for training and exporting models, making YOLO workflows more robust and user-friendly. 🚀
Overall, this update delivers a more stable, accurate, and user-friendly experience for anyone training, validating, or deploying YOLO models. 🎉
0.5 threshold by @WillieMaddox in https://github.com/ultralytics/ultralytics/pull/21165CustomizedValidator by @picsalex in https://github.com/ultralytics/ultralytics/pull/21196metrics summary method to val logs by @RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/21203nms=True by @Y-T-G in https://github.com/ultralytics/ultralytics/pull/21190generate_text_embeddings for DDP training by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/21210SolutionResults attributes in index.md and classwise_counts typo fix by @RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/21216lxml dependency by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/21221ultralytics 8.3.160 Clip keypoints for better visualization control by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/21220Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.3.159...v8.3.160
This release refactors and unifies the COCO evaluation process across detection, segmentation, and pose models, streamlining the codebase and improvin
This release refactors and unifies the COCO evaluation process across detection, segmentation, and pose models, streamlining the codebase and improving maintainability. It also brings enhancements to object counting, similarity search, NVIDIA Jetson benchmarking, and dependency management. 🛠️✨
coco_evaluate method now shared by detection, segmentation, and pose validators, reducing code duplication and ensuring consistent metric reporting.save_dir) is now included in detection model validation metrics, making experiment tracking and result management easier.save_dir in metrics helps users organize and reproduce results more efficiently.Overall, this release focuses on making the Ultralytics ecosystem more robust, user-friendly, and future-proof for both developers and end users. 🚀
show_in=False and show_out=False by @fn-hide in https://github.com/ultralytics/ultralytics/pull/21047save_dir to Metrics for better access by @RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/21136TextModel class for similarity search by @RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/21114ultralytics 8.3.159 pin model-compression-toolkit>=2.3.0,<2.4.1 by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/21161ultralytics 8.3.159 Refactor and Clean up COCO evaluation by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/21172Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.3.158...v8.3.159
This release streamlines image classification workflows and enhances the model selection experience in the Streamlit inference tool, making both more
This release streamlines image classification workflows and enhances the model selection experience in the Streamlit inference tool, making both more reliable and user-friendly. 🖼️🚀
Overall, this update delivers a smoother, more accurate, and user-friendly experience for both developers and end users working with Ultralytics models and tools.
Streamlit solution by @RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/21058ultralytics 8.3.158 Eliminate classification legacy transforms by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/21089Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.3.157...v8.3.158
Future-Proofing: Moves away from deprecated tools (pycocotools) to a modern, actively maintained alternative, ensuring long-term stability.
This release introduces much faster and more modern COCO/LVIS evaluation by integrating the faster-coco-eval library, along with several improvements for segmentation, pose, dataset handling, documentation, and dependency management. 🚀
pycocotools with the high-speed faster-coco-eval library for COCO and LVIS dataset evaluation, resulting in up to 4.5x faster validation. ⚡pycocotools) to a modern, actively maintained alternative, ensuring long-term stability.Overall, this update makes Ultralytics models faster, more reliable, and easier to use—whether you're training, validating, or just getting started! 🚀✨
flask>=3.0.1 for similarity search solution by @RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/21112verify_labels for GroundingDataset by @mohiuddin-khan-shiam in https://github.com/ultralytics/ultralytics/pull/21095macOS runners and pin OpenVINO>=2025.2.0 for macos-15 by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/20471ultralytics 8.3.157 Add faster-coco-eval package for COCO/LVIS evaluation by @MiXaiLL76 in https://github.com/ultralytics/ultralytics/pull/17020Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.3.156...v8.3.157
This release enhances model export reliability—especially for TensorRT INT8 quantization—improves data handling for training and calibration, and brin
This release enhances model export reliability—especially for TensorRT INT8 quantization—improves data handling for training and calibration, and brings documentation and example updates for a smoother user experience. 🚀
dynamic=True during INT8 export, making the process more flexible.drop_last option, letting users drop incomplete batches during data loading for both training and export workflows.requirements.txt for easier setup.This update is recommended for all users who export models, work with INT8 quantization, or want improved examples and documentation. 🚀🛠️
plot_images for classify training by @RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/21072ultralytics 8.3.156 Eliminate dynamic=True enforcement for TensorRT INT8 export by @Y-T-G in https://github.com/ultralytics/ultralytics/pull/20989Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.3.155...v8.3.156
This release (v8.3.155) delivers a key bug fix for YOLO-World models, improves developer experience with enhanced type annotations, and updates docume
This release (v8.3.155) delivers a key bug fix for YOLO-World models, improves developer experience with enhanced type annotations, and updates documentation and learning resources. 🚀📚
set_classes() method after running predictions, ensuring smoother workflow when updating classes dynamically.batch parameter supports both integers and floats, helping users leverage advanced batch size features.Overall, this update enhances reliability, developer experience, and learning resources for the Ultralytics community! 🌟
Solutions by @RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/21057ultralytics 8.3.155 Fix YOLO-World set_classes() error after prediction by @Y-T-G in https://github.com/ultralytics/ultralytics/pull/21051Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.3.154...v8.3.155
This release refactors and unifies the validation and metrics system across all YOLO tasks, making the codebase more modular, maintainable, and consis
This release refactors and unifies the validation and metrics system across all YOLO tasks, making the codebase more modular, maintainable, and consistent. It also brings UI improvements, performance boosts, and enhanced user feedback across the Ultralytics ecosystem. 🛠️✨
Major Validator & Metrics Refactor
Semantic Image Search UI Upgrade
Performance & Stability Improvements
.cpu() calls and adding caching for faster angle calculations.User Experience Enhancements
Bug Fixes
For Developers:
For End Users:
For the Community:
In summary:
This update makes Ultralytics' YOLO models and tools more powerful, user-friendly, and future-proof—whether you're building on the platform or using it for your projects! 🚀
similarity-search by @RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/21032.cpu() call and add @lru_cache by @RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/21022settings by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/21031imgsz for dynamic models by @Y-T-G in https://github.com/ultralytics/ultralytics/pull/21019model.fuse() to maintain NVIDIA JetPack 5 and below compatibility by @lakshanthad in https://github.com/ultralytics/ultralytics/pull/21028ultralytics 8.3.154 Refactor Validator and Metrics classes by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/21009Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.3.153...v8.3.154
This release introduces comprehensive OpenVINO benchmark results for YOLO11 models on the latest Intel® Core™ Ultra™ 7 265K hardware, alongside improv
This release introduces comprehensive OpenVINO benchmark results for YOLO11 models on the latest Intel® Core™ Ultra™ 7 265K hardware, alongside improved documentation, more accurate benchmarking data for Raspberry Pi and Rockchip devices, enhanced tracking examples, and a key fix for per-class metric reporting. 🚀
This update is especially valuable for developers and researchers deploying YOLO11 models on modern Intel chips or edge devices, and for anyone seeking reliable, easy-to-understand benchmarking and tracking workflows.
trackers README.md by @RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/20986rockchip-rknn.md by @RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/21006metrics.summary by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/20995ultralytics 8.3.153 OpenVINO COCO128 benchmarks by @ambitious-octopus in https://github.com/ultralytics/ultralytics/pull/20877Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.3.152...v8.3.153
Enhanced segmentation mask accuracy, improved memory management, faster confusion matrix processing, and better documentation for Ultralytics YOLO11.
Enhanced segmentation mask accuracy, improved memory management, faster confusion matrix processing, and better documentation for Ultralytics YOLO11. 🚀🖼️
Overall, this release delivers a smoother, more reliable, and more accessible experience for all Ultralytics YOLO11 users! 🎉
normalize modes for confusion matrix (≤30 classes) by @mihlefeld in https://github.com/ultralytics/ultralytics/pull/20955ultralytics 8.3.152 Optimize padding for Segment mask processing by @horsto in https://github.com/ultralytics/ultralytics/pull/20957Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.3.151...v8.3.152
This release brings cleaner, more customizable metric reporting across all tasks, improved tracking reliability, and efficiency boosts for video proce
This release brings cleaner, more customizable metric reporting across all tasks, improved tracking reliability, and efficiency boosts for video processing. It also enhances documentation and testing for a smoother user experience. 🚀✨
decimals and normalize options to all metric summary() methods, letting users control the precision and format of evaluation results.Overall, this update delivers a more polished, efficient, and user-friendly experience for anyone working with Ultralytics models and solutions!
solutions extras in uv export install by @RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/20956TrackZone by @RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/20928ultralytics 8.3.151 Add decimals argument to Metrics by @RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/20952Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.3.150...v8.3.151
This release streamlines model validation, improves memory efficiency for oriented bounding box (OBB) tasks, enhances documentation clarity, and fixes
This release streamlines model validation, improves memory efficiency for oriented bounding box (OBB) tasks, enhances documentation clarity, and fixes usability issues in the parking management solution. 🚀
pytorch-cpu) for Conda-based continuous integration tests, ensuring compatibility on machines without GPUs.Overall, this update delivers a smoother development and user experience, with optimizations that benefit both everyday users and contributors.
parking annotator by @RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/20926pytorch-cpu for Conda CI tests by @RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/20931ultralytics 8.3.150 Set default conf=0.01 for OBB validation by @Laughing-q in https://github.com/ultralytics/ultralytics/pull/20923Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.3.149...v8.3.150
This release enhances model export compatibility, improves usability for video and visualization workflows, and expands documentation and reporting fe
This release enhances model export compatibility, improves usability for video and visualization workflows, and expands documentation and reporting features across Ultralytics tools. 🚀
plot() method, ensuring all options are up-to-date and easy to find.✨ This update is all about making Ultralytics tools more robust, user-friendly, and ready for deployment across a wider range of platforms and workflows!
normalize and decimals support for confusion matrix export by @RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/20912visualization-args.md by @RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/20910yolo11n model support for IMX format in export table by @RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/20884waitKey() and destroyAllWindows() to Predictor by @Y-T-G in https://github.com/ultralytics/ultralytics/pull/20913ultralytics 8.3.149 Fix group-convolutions export bug for Edge TPU by @Y-T-G in https://github.com/ultralytics/ultralytics/pull/20919Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.3.148...v8.3.149
This release updates key dependencies and improves error handling, making model exports more reliable and troubleshooting easier for users. 🚀🔧
This release updates key dependencies and improves error handling, making model exports more reliable and troubleshooting easier for users. 🚀🔧
onnxslim dependency to version 0.1.56 for ONNX and TensorFlow SavedModel exports, ensuring compatibility with the latest features and bug fixes.numpy._core error while loading cache by @Y-T-G in https://github.com/ultralytics/ultralytics/pull/20904ultralytics 8.3.148 update onnxslim>=0.1.56 dependency by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/20905Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.3.147...v8.3.148
One column per quarter.