A desktop labeler that swallowed the entire CV model zoo

What it does

X-AnyLabeling is a desktop annotation application that wraps a staggering number of computer-vision models behind a unified interface. It aims to replace the usual patchwork of scripts by handling images and videos in one place, then exporting to formats like COCO, YOLO, VOC, DOTA, and ShareGPT.

The interesting bit

The project doesn’t just run YOLO and SAM; it also integrates vision-language models (Qwen3-VL, Gemini, Florence2), chatbots, document parsing, video classification timelines, and 3D cuboid annotation. Inference can run locally via ONNX Runtime, TensorRT, or OpenCV DNN, or be offloaded to a remote server if your GPU lives elsewhere.

Key highlights

• Model zoo spans detection (YOLOv5 through YOLO26), segmentation (SAM 1/2/3, EdgeSAM), pose estimation, OCR (PaddleOCR), object counting, grounding, and VLMs.
• Annotation types cover polygons, rotated boxes, cuboids, lines, keypoints, and text detection / recognition / KIE.
• Supports one-click batch inference across an entire task, plus annotation review workflows.
• UI localized in English, Chinese, Japanese, and Korean.

Caveats

• The README notes a recent PyQt5-to-PyQt6 refactor is in Beta, so the UI layer may still be settling.
• The license badge reads LGPL v3 while the project text states GPL-3.0; the actual terms are unclear at a glance.

Verdict

Ideal for data engineers and researchers who need industrial-scale, multi-modal labeling without chaining together half a dozen tools. If your needs stop at manual bounding boxes, this is massive overkill.