ultralytics 8.3.230

v8.3.230 - `ultralytics 8.3.230` Fix color conversion for Tensor inputs and PIL plots (#22734)

on Python PyPI

🌟 Summary

ultralytics 8.3.230 is a stability and UX-focused release that fixes color issues in visualizations, improves TensorRT and device handling, and loosens dependency constraints for smoother installs, along with some docs and CI polish. 🎨⚙️🚀

📊 Key Changes

• 🎨 Accurate colors for visualizations (Tensor + PIL) [@Y-T-G, #22734]

  • Fixes channel order for tensor inputs in predictors (RT-DETR, SAM, YOLO detect, YOLO classify).
  • Ensures pil=True plots return proper RGB images instead of visually “off” colors.
  • Results.plot() now delegates to annotator.result(pil) for consistent behavior.

• ⚙️ More robust TensorRT dynamic input handling [@Laughing-q, #22729]

  • For dynamic TensorRT models, the input shape now uses the maximum profile shape, not the “optimal” one.
  • Reduces runtime shape errors when running YOLO models at larger or varying resolutions.

• 🧠 Safer device selection for CPU / MPS [@Y-T-G, #22740]

  • When using select_device("cpu") or select_device("mps"), CUDA_VISIBLE_DEVICES is now set to "" instead of "-1".
  • Avoids crashes and odd CUDA behavior on some systems (including Windows via Ultralytics HUB).

• 📦 Looser dependency version pins for better compatibility [@glenn-jocher, #22721]

  • Removes upper bounds on major dependencies: numpy, matplotlib, opencv-python, pillow, pyyaml, requests, scipy, torchvision, psutil, polars.
  • PyTorch requirements updated to:
    • General: torch>=1.8.0
    • Windows: torch>=1.8.0,!=2.4.0 (skips known-bad 2.4.0 on Windows CPU).
  • ultralytics-thop now uses >=2.0.18 instead of an upper cap.

• 🖼️ Docs UX + SEO improvements [@sergiuwaxmann, @glenn-jocher, @RizwanMunawar]

  • Docs banner updated to highlight Ultralytics’ $30M Series A with new visuals and investor logos.
  • Page title logic made safer to handle pages without TOC or page metadata, preventing template errors.
  • Object blurring & cropping guides now embed a new, clearer YouTube tutorial (J1BaCqytBmA), aligned with current YOLO11 workflows.

• 🧪 CI & workflow maintenance [@dependabot[bot]]

  • All GitHub workflows now use actions/checkout@v6 (including CI, docs, docker, link checks, publishing, mirroring).
  • Keeps automation aligned with the latest GitHub Actions best practices.

🎯 Purpose & Impact

• 🎨 More trustworthy visual outputs

  • Fixes incorrect RGB/BGR handling so prediction overlays look right across RT-DETR, SAM, YOLO detect, and YOLO classify.
  • Especially important for users working directly with tensors or using pil=True for plotting or UI integrations.

• 🧩 Smoother deployment on TensorRT and varied input sizes

  • Using the maximum TensorRT profile shape for dynamic inputs reduces shape mismatches and runtime failures.
  • Helpful for production setups that need to handle different image resolutions reliably.

• 💻 More reliable CPU/MPS runs and Ultralytics HUB integration

  • Clearing CUDA_VISIBLE_DEVICES properly avoids crashes and confusing CUDA behavior on systems where GPUs must be “hidden.”
  • Leads to more predictable device selection across CLI, Python API, and Ultralytics HUB.

• 📦 Easier installs and upgrades

  • Looser dependency pins mean fewer version conflicts with existing environments (e.g., modern numpy, torch, torchvision).
  • Still protects Windows users from known PyTorch 2.4.0 CPU issues.

• 📚 Better docs experience & onboarding

  • Updated banner and safer HTML title logic improve stability, branding, and SEO for documentation.
  • New video tutorials make object blurring/cropping with YOLO11 easier to learn and replicate.

• 🏗️ More robust CI/CD

  • Upgrading to actions/checkout@v6 keeps the project’s automation current, aiding long-term reliability of builds, tests, and releases.

Overall, v8.3.230 is a quality-focused release: if you visualize results, run dynamic TensorRT models, rely on CPU/MPS, or manage modern Python stacks, you should see more stable and predictable behavior. ✅

What's Changed

• Revert "fix(pyproject.toml): 🐞 Update dependency versions in pyproject.toml to enforce upper limits" by @glenn-jocher in #22721
• Fix validation with dynamic TensorRT model by setting the max_shape as input_shape by @Laughing-q in #22729
• Update Banner by @sergiuwaxmann in #22689
• Bump actions/checkout from 5 to 6 in /.github/workflows by @dependabot[bot] in #22741
• Update main.html to zensical compat by @glenn-jocher in #22746
• Add https://youtu.be/J1BaCqytBmA to docs by @RizwanMunawar in #22745
• Set CUDA_VISIBLE_DEVICES to empty string instead of -1 to avoid Windows crash by @Y-T-G in #22740
• ultralytics 8.3.230 Fix color conversion for Tensor inputs and PIL plots by @Y-T-G in #22734

Full Changelog: v8.3.229...v8.3.230