pypi ultralytics 8.3.229
v8.3.229 - `ultralytics 8.3.229` 300% faster COCO Segmentation val (#22651)
on Python PyPI

10 hours ago

🌟 Summary

YOLO 8.3.229 focuses on much faster COCO instance segmentation validation (~3×), tighter dependency and environment controls, improved multi‑GPU behavior, and a broad documentation refresh (Jetson, datasets, guides) for smoother real‑world use. 🚀🧠

📊 Key Changes

  • ⚡ 300% Faster COCO Segmentation Validation (PR #22651)

    • Replaces threaded faster_coco_eval RLE encoding with a native, optimized RLE pipeline.
    • Cleans up mask scaling by using a more efficient ops.scale_masks API with ratio_pad support.
    • Simplifies segmentation validation imports by removing per‑mask thread pools.

  • 🔐 Safer, More Predictable Environments

    • Dependency upper bounds in pyproject.toml (PR #22701) to avoid breaking changes in numpy, torch, opencv, pillow, polars, etc.
    • Platform‑specific torch constraints to avoid known Windows issues.
    • IMX export guards (PR #22673): explicitly blocks unsupported python=3.12 and torch>=2.9 with clear errors.
    • Official Docker image updated to PyTorch 2.9.1 (CUDA 12.8, cuDNN 9) (PR #22708), and build simplified by relying on uv from the base image (PR #22709).

  • 🖥️ More Robust Multi‑GPU & CI

    • GPU CI re‑enabled with updated tests, including larger coco128 training for better multi‑GPU validation (PR #22710).
    • Distributed sampler fix when batch_size >= dataset_size (PR #22714): safely falls back to batch_size=1 to avoid failures on tiny datasets.
    • Clearer GPU auto‑selection logs & warnings when fewer idle GPUs than requested are available (PR #22716).
    • GitHub Actions runner image bumped to 2.329.0 (PR #22707).

  • 📚 Major Documentation & Guide Updates

    • Massive docs cleanup & refresh (PR #22698):
      • Consistent YOLO11 branding, AGPL links, and licensing text.
      • Clearer setup guides for Docker, AzureML, Jetson, Raspberry Pi, DeepStream, Coral Edge TPU, Vertex AI, etc.
      • Improved dataset docs (MNIST, COCO variants, DOTA, pose & segment subsets, Explorer, tracking) with concrete structures and examples.
    • Jetson guide updated for Jetson AGX Thor (JetPack 7) with specs, flashing steps, and extended benchmarks (PR #22699).
    • Numerous small wording/typo and link fixes across guides, tasks, integrations, and macros to match current APIs and best practices.

  • 📖 Docs Build & Chat Widget Improvements

    • HTML/JS/CSS minification made safer and more efficient, with preservation of script, pre, code, textarea content and regex‑based HTML cleanups (PRs #22676, #22678).
    • Local docs server UX improved with clearer logging and error handling (PR #22676).
    • Docs chat widget upgraded stepwise to Ultralytics LLM chat.js v0.0.6 with separate chatExamples and searchExamples for better in‑docs help (PRs #22680, #22691, #22696).
    • Fixes to a quoting bug that could break the chat widget and minor Explorer API doc cleanups (PRs #22663, #22666).

  • 🎨 Repo & Config Quality

    • Prettier print-width updated to 120 and applied across docs JS/HTML/CSS plus dataset YAML for cleaner diffs and readability (PR #22706).
    • Many argument tables/macros (train, val, export, predict, solutions, visualization) tightened up for clarity and consistency.

  • 🎥 & Learning Experience

    • KITTI dataset docs now embed a step‑by‑step YOLO11 training video (PR #22685), making it easier for newcomers to follow along.

🎯 Purpose & Impact

  • Faster Experiment Cycles for Segmentation

    • The new COCO segmentation validation path removes per‑mask threading overhead and uses an in‑repo RLE encoder and improved mask scaling, making val runs much faster and more consistent.
    • This directly benefits users iterating on YOLO11‑seg models, especially when validating frequently on COCO‑style datasets. 🏎️

  • More Stable, Reproducible Installs

    • Upper‑bounded dependencies and IMX‑specific version checks reduce “works yesterday, broken today” failures from upstream library changes.
    • Clear errors around unsupported Python/PyTorch combos save time debugging export failures, especially on specialized platforms like Sony IMX. 🛡️

  • More Reliable Multi‑GPU Training & Testing

    • The sampler fix and GPU logging improvements help avoid silent edge‑case bugs when using small datasets or requesting more GPUs than truly idle.
    • Re‑enabled GPU CI with realistic tests increases confidence that multi‑GPU behavior stays correct in future releases. ✅

  • Better Docs for Real Deployments

    • Expanded Jetson coverage (including AGX Thor/JetPack 7 and updated benchmarks) plus refreshed deployment guides (AzureML, DeepStream, Vertex AI, etc.) make it easier to move models from prototype to production on modern hardware.
    • Clearer dataset docs, label formats, and examples lower the barrier for users building their own detection, segmentation, pose, OBB, and tracking pipelines. 📘

  • Smoother Documentation UX

    • Safer minification and chat widget upgrades help ensure docs load reliably, scripts don’t break, and users can quickly discover relevant help via chat and search suggestions. 💬

Overall, v8.3.229 is a performance‑ and robustness‑oriented release: your COCO segmentation validation is faster, environments are safer, multi‑GPU workflows are more predictable, and the documentation better reflects how to run YOLO11 in real systems. 🌐🔥

What's Changed

Full Changelog: v8.3.228...v8.3.229

Check out latest releases or
releases around ultralytics 8.3.229

Don't miss a new ultralytics release

NewReleases is sending notifications on new releases.

Get notifications