v2.3.1

CVHub520/X-AnyLabelingv2.3.1Jan 31, 2024by CVHub520

AI Summary

This is a minor release (v2.3.1) of X-AnyLabeling, an AI-powered labeling tool. The update introduces three key enhancements: the ability to save cropped rectangle shapes, a new CLIP+SAM model combination for improved semantic and spatial understanding, and support for the Depth Anything model in depth estimation tasks.

Key Highlights

Added support for saving cropped rectangle shapes
Combined CLIP and SAM models for enhanced semantic and spatial understanding (EdgeSAM with Chinese CLIP)
Introduced Depth Anything model for depth estimation task
GPU version compiled with CUDA 11.6 (may have compatibility issues with other CUDA versions)
Expanded support for 30+ models across various computer vision tasks

New Features

Save cropped rectangle shapes functionality
CLIP and SAM model combination for auto-labeling (EdgeSAM with Chinese CLIP)
Depth Anything model support for depth estimation

Full Release Notes

## Released versions

| Version | MD5                                |
|---------|------------------------------------|
| X-AnyLabeling-CPU.exe | 73e5d39cb23df58004b2c795f54e88b9 |
| X-AnyLabeling-GPU.exe | 745dc17bf9996a0b5f8e00e3ffe5f7df |
| X-Anylabeling-Linux-CPU | f1d1390906c1081aa5ad9aec6605f062 |
| X-Anylabeling-Linux-GPU | c281893cfc2e61081fdbf8bb6b20ef96 |

> Note: The GPU version mentioned above is compiled based on CUDA 11.6. Different versions may result in issues with xxx functionality. It is recommended to consider running the application using the source code, especially if compatibility issues arise.

## Supported models

| Task                     | Model                  |
|--------------------------|------------------------|
| **Image Classification** | ResNet50, InternImage, YOLOv5-cls, YOLOv8-cls, PULC Person/Vehicle Attribute |
| **Object Detection**     | YOLOv5, YOLOv6, YOLOv7, YOLOv8, YOLOX, YOLO-NAS, DAMO_YOLO, GOLD_YOLO, RT-DETR, RTMDet |
| **Instance Segmentation** | YOLOv5-seg, YOLOv8-seg |
| **Keypoint Detection**   | RTMPose, DWPose, YOLOv6-face, YOLOv8-pose |
| **Oriented Object Detection** | YOLOv5_obb, YOLOv8_obb        |
| **Multi-Object Tracking** | OC-Sort, ByteTrack   |
| **Segment Anything**     | SAM, SAM-HQ, EdgeSAM, MobileSAM, EfficientViT-SAM, Med-SAM2D |
| **Optical Character Recognition**                  | PPOCRv4                |
| **Land Detection**       | CLRNet                 |
| **Image Captioning**     | RAM                    |
| **Visual Language Model** | Grounding-DINO, Chinese-CLIP    |
| **Depth Estimation** | Depth Anything |

## Important Updates

  - ✅ Support save cropped rectangle shapes.
  - ✅ Combining CLIP and SAM models for enhanced semantic and spatial understanding. An example can be found [here](./anylabeling/configs/auto_labeling/edge_sam_with_chinese_clip.yaml).
  - ✅ Adding support for the [Depth Anything](https://github.com/LiheYoung/Depth-Anything.git) model in the depth estimation task.