Starstruk Logo MarkStarstruk

v4.0.0-beta.8

CVHub520/X-AnyLabelingv4.0.0-beta.8Jun 14, 2026by CVHub520

AI Summary

This release introduces AI-assisted video annotation workflows, vertex eraser tools, and temporary panning features, while fixing stability issues regarding GPU VRAM management, object selection, and various model integrations.

Key Highlights

  • Add an AI-assisted video annotation workflow
  • Add vertex eraser support for polygon and line strip annotations
  • Support temporary Space-drag panning while drawing shapes
  • Add PaddleOCR-VL-1.6 API model support
  • Prevent SAM2 interactive segmentation from leaking GPU VRAM

New Features

  • AI-assisted video annotation workflow
  • Vertex eraser support for polygons and lines
  • Temporary Space-drag panning while drawing
  • PaddleOCR-VL-1.6 API model support
  • Improved Chatbot vision capability detection

Full Release Notes

https://github.com/user-attachments/assets/33a57390-683d-4a24-b0cf-3668af5f7a13

> PyPI: https://pypi.org/project/x-anylabeling-cvhub/
> Baidu Cloud: https://pan.baidu.com/s/1pgaw02inCvbEgOme9ajDJA?pwd=e528

> [!NOTE]
> Due to compatibility issues across different systems, if the precompiled version doesn’t work properly on your machine, you can try building and running it from source instead. For details, check out the official installation guide and user documentation.

### 🚀 New Features

- Add an AI-assisted video annotation workflow to the video classifier. (#1224,#1330)
- Add vertex eraser support for polygon and line strip annotations. (#892)
- Support temporary Space-drag panning while drawing shapes. (#1368)
- Add PaddleOCR-VL-1.6 API model support.
- Improve Chatbot vision capability detection for supported models.

### 🐛 Bug Fixes

- Stabilize object selection and label rendering in the label list. (#1385)
- Preserve group IDs after auto-labeling results are applied. (#1376)
- Prevent SAM2 interactive segmentation from leaking GPU VRAM during CUDA inference.
- Match auto-labeling model dropdown searches against display names. (#1379)
- Skip partially out-of-bounds OBB exports in label conversion. (#1383)
- Pass all Ultralytics training parameters while excluding internal-only X-AnyLabeling options.
- Align ONNX dependency checks with the package requirements to avoid false missing-package errors.
- Surface Anthropic model fetch errors in Chatbot and honor request timeouts.
- Restore Chatbot model refresh behavior and improve dark mode controls.

### 🛠️ Improvements

- Add LocateAnything grounding documentation and references. (#1382)
- Clarify Rex-Omni GPU selection behavior for the vLLM backend.
- Add bottom spacing to the progress dialog cancel button.

### 🌟 Contributors

A total of 4 developers contributed to this release.

Thank @GIGIGIGIbaby, @Snag9311, @Shuwei Ji,  @CVHub520


**Full Changelog**: https://github.com/CVHub520/X-AnyLabeling/compare/v4.0.0-beta.7...v4.0.0-beta.8