ultralytics 8.3.239

v8.3.239 - SAM3 and `SystemLogger()` improvements (#22965)

on Python PyPI

🌟 Summary

Ultralytics v8.3.239 focuses on smoother SAM 3 text prompting, richer system monitoring via SystemLogger(), quieter logs, and broader ONNX export support (including Jetson JetPack 6), plus improved contributor attribution in the docs. 🚀

📊 Key Changes

• SAM 3 text prompting made plug‑and‑play 🧩

  • Replaced the custom SAM3 tokenizer with CLIP’s built‑in SimpleTokenizer().
  • Removed the need to manually download and pass bpe_simple_vocab_16e6.txt.gz.
  • SAM3SemanticPredictor and SAM3VideoSemanticPredictor no longer take a bpe_path argument.
  • If clip isn’t installed, Ultralytics auto-installs it from the Ultralytics CLIP repo.
  • Updated SAM 3 docs and reference nav to reflect the simpler API and removed tokenizer module.

• SystemLogger() gains real‑time throughput metrics 📈

  • SystemLogger.get_metrics() now supports rates=True to report disk and network MB/s:
    • Disk: read_mbs, write_mbs
    • Network: recv_mbs, sent_mbs
  • Default behavior (rates=False) still returns cumulative MB since logger init.
  • Adds internal tracking of previous disk/net stats and time for safe rate calculations.
  • Example docs updated to show GPU keys as strings ("0", "1").

• Centralized warning suppression for cleaner logs 🔕

  • Moved scattered warnings.filterwarnings(...) calls into ultralytics/utils/__init__.py.
  • Removed local warning suppression from:
    • Exporter (ultralytics/engine/exporter.py)
    • MobileCLIP text model loader (ultralytics/nn/text_model.py)
    • Plotting utils (ultralytics/utils/plotting.py)
    • TensorBoard callbacks (ultralytics/utils/callbacks/tensorboard.py)
  • TensorBoard graph logging now:
    • Tries a simple trace first, then an RT-DETR‑style export fallback.
    • Logs a single, more informative warning if both fail.

• ONNX export updated to support ONNX 1.20.0+ 📦

  • Exporter now requires onnx>=1.12.0,<2.0.0 instead of capping at 1.19.1 for:
    • export_onnx
    • export_saved_model (TensorFlow route)
  • Jetson JetPack 6 Dockerfile no longer hard‑pins ONNX below 1.20.0; the workaround was removed.
  • Comment in pyproject.toml updated to note MacOS testing up to onnx==1.20.0.

• Docs & authorship improvements 📚

  • SAM 3 docs simplified: all examples now construct SAM3SemanticPredictor / SAM3VideoSemanticPredictor without bpe_path.
  • mkdocs.yml nav updated to drop the removed tokenizer_ve reference page.
  • docs/mkdocs_github_authors.yaml expanded with many new GitHub usernames and avatars, plus aliases and Ultralytics team mappings for better contributor attribution.

• CI / tooling maintenance 🧰

  • GitHub Actions: actions/download-artifact bumped from v6 → v7 (Node.js 24 runtime) in publish.yml.
  • Package version bumped to 8.3.239 in ultralytics/__init__.py for clear release tracking.

🎯 Purpose & Impact

• Easier SAM 3 text-based segmentation 💬

  • No more manual BPE vocabulary downloads or bpe_path wiring.
  • Using CLIP’s own tokenizer reduces duplicated code and aligns SAM 3 text prompts with standard CLIP tooling.
  • SAM 3 examples become more “drop‑in” for new users: just provide sam3.pt and go.

• Better monitoring for training and deployment 📊

  • SystemLogger.get_metrics(rates=True) lets you watch live I/O throughput, which is especially helpful when:
    • Diagnosing dataset loading bottlenecks.
    • Monitoring training jobs on shared servers or in production.
  • Preserving the old cumulative behavior avoids breaking existing dashboards or scripts.

• Quieter, more focused logs 🤫

  • Centralized warning suppression cuts repeated, noisy framework warnings (timm, CoreML, ONNX/TorchScript, matplotlib) across the codebase.
  • Makes logs easier to read, especially during long training/export runs or when using TensorBoard.

• More robust ONNX export, including Jetson 🖥️

  • Supporting ONNX up to <2.0.0 (including 1.20.0) unblocks environments like Jetson JetPack 6 that depend on newer ONNX.
  • Removing the Jetson Dockerfile hack avoids surprising dependency rewrites and simplifies maintenance.

• Improved documentation and community recognition 🤝

  • Cleaner, up‑to‑date docs around SAM 3 reduce setup friction and confusion.
  • Expanded author mapping ensures more contributors are correctly credited on documentation pages, which encourages community engagement.

Overall, v8.3.239 is a quality‑of‑life release: SAM 3 is simpler to use, monitoring is more informative, logs are tidier, and ONNX export is more future‑proof—without breaking existing workflows. ✅

What's Changed

• Restore GitHub authors with avatars and usernames by @glenn-jocher in #22961
• feat: 📦 update onnx version to 1.20.0 by @onuralpszr in #22867
• Bump actions/download-artifact from 6 to 7 in /.github/workflows by @dependabot[bot] in #22951
• Use CLIP tokenizer functionality and remove the need of bpe_simple_vocab for SAM3 by @Y-T-G in #22958
• Centralize warning suppression by @glenn-jocher in #22962
• SystemLogger rate mode by @glenn-jocher in #22964
• SAM3 and SystemLogger() improvements by @glenn-jocher in #22965

Full Changelog: v8.3.238...v8.3.239