fix(deps): update dependency ultralytics ( 8.3.241 → 8.3.243 ) - #240

v8.3.243: - ultralytics 8.3.243 Deduplicate ConsoleLogger progress bars (#​23082)

Compare Source

🌟 Summary (single-line synopsis)

v8.3.243 sharpens Ultralytics training telemetry and console output—cleaner progress bars, richer run metadata for Ultralytics Platform/HUB-style tracking, and safer/less noisy logging 🚀🧹📡

📊 Key Changes

• (Priority) ConsoleLogger progress-bar deduplication + batching 🧾✅
  Suppresses noisy in-progress bar updates (keeps completion signals) and prevents repeated “100%” lines for the same step (epoch/train/val/Class). Also adds batched flushing (by lines/time) and safer handler cleanup.
• Richer “training_started” metadata for platform telemetry 🧠🌍
  Adds a new _get_environment_info() collector (OS, Python, hostname, CPU/GPU details, command, and Git repo/branch/commit when available) plus model info (params, GFLOPs, class count) into the training-start event payload.
• Checkpoint upload timing tweak (rate-limit initialization) ⏱️⬆️
  Initializes the upload timer at training start so the first checkpoint upload waits ~15 minutes, reducing immediate upload churn at the beginning of training.
• Headless install docs promoted 🖥️🚫
  Documentation now recommends pip install ultralytics-opencv-headless for servers/CI/Docker to avoid OpenCV GUI libGL issues.
• Docker GPU run instructions clarified 🐳⚡
  Example docker run commands now include --runtime=nvidia to reduce “GPU not found” surprises.
• Opt-out of auto requirements checks 🧰🛑
  New env var ULTRALYTICS_SKIP_REQUIREMENTS_CHECKS=1 skips requirement checks/auto-install behavior (useful for controlled environments).
• Small correctness/docs fixes 📝
  Pose loss comments corrected, and YOLO-World docs note a community tip: adding "" as a background class may help in some cases.

🎯 Purpose & Impact

• Cleaner training logs (especially in notebooks/CI/streamed consoles) 🧹
  Less spammy progress output makes it easier to spot real warnings, metrics, and errors—especially when logs are forwarded to external systems.
• Better experiment traceability and reproducibility 🔍
  Environment + Git + model metadata makes runs easier to compare and debug across machines, containers, and branches.
• More predictable remote syncing ⏱️
  Delaying the first checkpoint upload reduces bandwidth spikes early in training and avoids uploading “too-early” checkpoints.
• Smoother installs on servers and pipelines 🧩
  Headless package guidance + optional skipping of requirements checks helps teams run in Docker/CI/locked-down environments with fewer dependency headaches.
• No major model architecture changes ✅
  This release is primarily about logging/telemetry, usability, and docs, rather than changing YOLO model behavior.

What's Changed

• Headless docs by @​glenn-jocher in #​23077
• Update callbacks by @​glenn-jocher in #​23079
• Add missing --runtime=nvidia usage description to Dockerfiles by @​lakshanthad in #​23080
• Fix incorrect loss order comments in v8PoseLoss by @​raimbekovm in #​23060
• Add note about background class for YOLO-World models by @​raimbekovm in #​23058
• Update pyproject.toml to zensical>=0.0.15 by @​glenn-jocher in #​23052
• feat: ✨ add option to skip requirements checks via environment variable by @​onuralpszr in #​23065
• ultralytics 8.3.243 Deduplicate ConsoleLogger progress bars by @​glenn-jocher in #​23082

Full Changelog: ultralytics/ultralytics@v8.3.242...v8.3.243

v8.3.242: - ultralytics 8.3.242 New ultralytics-opencv-headless PyPI package (#​23075)

Compare Source

🌟 Summary (single-line synopsis)

Ultralytics 8.3.242 adds a new headless OpenCV distribution (ultralytics-opencv-headless) to make installs on servers/CI/containers smoother and more reliable 📦🖥️

📊 Key Changes

• 🆕 New PyPI package: ultralytics-opencv-headless (PR #​23075 by @​glenn-jocher)
  • Publishes a second package alongside ultralytics, swapping opencv-python → opencv-python-headless during the build.
• 🐳 Jetson/Docker dependency fixes
  • JetPack 5 Docker: pins NumPy to 1.23.5 to avoid TensorRT export breakage (PR #​23032) 🔧
  • JetPack 6 + Python export Docker: removes unnecessary NumPy pinning to reduce conflicts (PR #​23030) 🧹
• 🧩 Segmentation utility improvement
  • masks2segments() now accepts NumPy arrays or PyTorch tensors, with safer dtype/contiguity handling for OpenCV contours (PR #​23025) ✅
• 📤 Export robustness improvements (Axelera + INT8 calibration)
  • Adds fraction to Axelera export args and ensures a default calibration dataset when missing (PR #​23045) ⚙️
• 🧭 Training correctness fix for Oriented Bounding Boxes (OBB)
  • Fixes swapped width/height scaling in v8OBBLoss filtering of tiny rotated boxes (PR #​23061) 🎯
• 🧾 Docs & cleanup
  • Docker outputs persistence guidance, SAM 3 doc refresh + embedded tutorial, hyperparameter tuning docs updated (dfl), plus small code cleanups/staticmethod refactors (multiple PRs) 📚✨

🎯 Purpose & Impact

• 🚀 Easier installs in headless environments: ultralytics-opencv-headless avoids GUI-related OpenCV dependencies, reducing common install/runtime issues on Linux servers, Docker, and CI.
• 📉 Lean(er) deployments: headless OpenCV often means fewer system GUI libs and less friction when containerizing.
• 🧪 More robust post-processing: masks2segments() works cleanly with both NumPy and Torch pipelines, with fewer OpenCV edge-case failures.
• 🐳 More reliable Jetson export workflows: clearer, safer NumPy handling across JetPack 5 vs JetPack 6/export images reduces “dependency tug-of-war”.
• 🧭 Improved OBB training stability: correct image dimension scaling helps prevent wrongly filtering valid rotated labels, potentially improving convergence and quality for rotated-box tasks.

If you want, I can also suggest when to install ultralytics vs ultralytics-opencv-headless based on your environment (desktop vs Docker/CI).

What's Changed

• Update masks2segments to support numpy input by @​Laughing-q in #​23025
• Fix CIB docstrings in block.py (#​18585) by @​raimbekovm in #​23024
• Pin numpy for JetPack 5 Docker builds by @​lakshanthad in #​23032
• Remove numpy pin for Sony IMX export and JetPack 6 systems by @​lakshanthad in #​23030
• Clarify Docker output directory in Quickstart Guide (fixes #​22938) by @​raimbekovm in #​23029
• Remove redundant assert statements in loss.py by @​raimbekovm in #​23028
• Update sam-3.md docs by @​Laughing-q in #​23041
• Add https://youtu.be/BE2Nu3edyOo to docs by @​RizwanMunawar in #​23037
• fix: correct loss name from box_loss to cls_loss in TVPDetectLoss by @​ShuaiLYU in #​23046
• Capitalize markdown by @​glenn-jocher in #​23048
• docs: update hyperparameter tuning search space by @​raimbekovm in #​23049
• Update docs and minor code optimization by @​Laughing-q in #​23045
• Fix incorrect width/height indexing in v8OBBLoss by @​raimbekovm in #​23061
• Fix incorrect return type in _get_covariance_matrix docstring by @​raimbekovm in #​23062
• Fix misleading comment in v8OBBLoss.call() by @​raimbekovm in #​23063
• ultralytics 8.3.242 New ultralytics-opencv-headless PyPI package by @​glenn-jocher in #​23075

New Contributors

• @​raimbekovm made their first contribution in #​23024

Full Changelog: ultralytics/ultralytics@v8.3.241...v8.3.242