Starstruk Logo MarkStarstruk

v3.0.0

CVHub520/X-AnyLabelingv3.0.0May 15, 2025by CVHub520

AI Summary

X-AnyLabeling v3.0.0 is a major release introducing Visual Prompting for detection and segmentation, a new Chatbot feature for annotating multimodal datasets for Vision-Language Models, and support for multiple new models including YOLO12, D-FINE, RF-DETR, and GeCo zero-shot counting.

Key Highlights

  • Introduced Visual Prompting for Detection and Segmentation - a new AI-powered annotation approach
  • Added Chatbot feature for annotating multimodal datasets for Vision-Language Models (VLMs)
  • Support for new models: YOLO12, D-FINE, RF-DETR, Grounding-DINO-1.6-API, and GeCo zero-shot counting
  • Enabled model downloading from ModelScope in addition to existing sources
  • Added ffmpeg acceleration and non-ASCII path support for better compatibility

Breaking Changes

  • Removed imgviz dependency from requirements - users relying on imgviz-specific functionality may need to adjust their workflows

New Features

  • Visual Prompting for Detection and Segmentation
  • Chatbot for multimodal dataset annotation for VLMs
  • ffmpeg acceleration and non-ASCII path support (#891)
  • ModelScope model downloading integration
  • One-click import/export for VLM-R1-OVD labels
  • Automatic saving of group IDs when using G (group) and U (ungroup) shortcuts
  • GroupID filter and improved label filtering functionality
  • Support for GeCo zero-shot counting model
  • Support for Grounding-DINO-1.6-API open-set detection
  • Support for YOLO12 object detection
  • Support for D-FINE object detection (with ONNX exporter)
  • Support for RF-DETR object detection (with ONNX exporter)

Full Release Notes

https://github.com/user-attachments/assets/f517fa94-c49c-4f05-864e-96b34f592079

<br>

<img src="https://github.com/user-attachments/assets/7f43bcec-96fd-48d1-bd36-9e5a440a66f6" width="100%" />
<div align="center"><strong>Visual Prompting for Detection</strong></div>

<br>

<img src="https://github.com/user-attachments/assets/208dc9ed-b8c9-4127-9e5b-e76f53892f03" width="100%" />
<div align="center"><strong>Visual Prompting for Segmentation</strong></div>

<br>

<img src="https://github.com/user-attachments/assets/56c9a20b-c836-47aa-8b54-bad5bb99b735" width="100%" />
<div align="center"><strong>Chatbot</strong></div>

### 🚀 New Features

- Add ffmpeg acceleration and non-ASCII path support (#891)
- Allow downloading models from [ModelScope](https://www.modelscope.cn/collections/X-AnyLabeling-7b0e1798bcda43) in addition to existing sources
- Enable one-click import and export of labels for [VLM-R1-OVD](https://github.com/om-ai-lab/VLM-R1)
- Enable the [Chatbot](./docs/en/chatbot.md) to annotate multimodal datasets for Vision-Language Models (VLMs)
- Enable automatic saving of group IDs when grouping or ungrouping shapes using shortcut keys G (group) and U (ungroup). (#855)
- Introduce GroupID filter and improve label filtering functionality (#686)
- Support [GeCo](./examples/counting/geco/README.md) zero-shot counting model (#863)
- Support [Grounding-DINO-1.6-API](https://algos.deepdataspace.com/en#/model/grounding_dino) open-set object detection model
- Support [YOLO12](https://arxiv.org/abs/2502.12524) object detection model
- Support [D-FINE](./tools/onnx_exporter/export_dfine_onnx.py) object detection model
- Support [RF-DETR](./tools/onnx_exporter/export_rfdetr_onnx.py) object detection model

### 🐛 Bug Fixes

- Fix bug in `predict_shapes` for `Florence-2` model (#913)
- Replace `os_sorted` with `natsorted` to avoid potential segfault (#906)
- Merge multi-part segmentations into single instance on export (#910)
- Handle exceptions in model loading by initializing local_model_data to an empty dictionary for improved stability (#901)
- Prevent UI disappearance when ESC key is pressed during AI annotation (#423)
- Fixed the bug about exporting the empty labels (#881)

### 🛠️ Improvements

- Move image conversion to avoid redundant processing when using cached embeddings (#915)
- Add new FAQs addressing common runtime errors and file loading issues, including solutions and references to related GitHub issues (#869, #906, #907)
- Enhance image processing logic to support dynamic batch handling based on model type, improving efficiency in auto-labeling operations
- Introduce `iou_threshold` and `conf_threshold` parameters across various model configurations for enhanced detection accuracy
- Remove imgviz dependency from requirements and update colormap implementation in labeling utilities for improved modularity
- Optimize batch processing with UI/backend separation (#757)
- Enhance shape visibility handling in labeling interface (#669)
- Updated the merge_shapes method to handle both rectangle and polygon shapes, allowing for more versatile shape unions. (#561)

### 🌟 Contributors

A total of 8 developers contributed to this release.

Thank @Pecako2001, @liutao, @shyhyawJou, @talebolano, @urbaneman, @wangxiang0722, @Little-King2022, @CVHub520

**Full Changelog**: https://github.com/CVHub520/X-AnyLabeling/compare/v2.5.4...v3.0.0