CVHub520/X-AnyLabeling

X‑AnyLabeling – AI‑assisted annotation desktop app

What it is – A cross‑platform GUI tool that lets you label images, video, text, and 3D point clouds. It bundles a large collection of state‑of‑the‑art vision models (YOLO, SAM, Grounding‑DINO, etc.) so you can auto‑generate annotations and then edit them manually.

Why it matters – High‑quality labeled data is the bottleneck for training computer‑vision models. X‑AnyLabeling speeds up that process by:

• Providing ready‑made annotation primitives (boxes, polygons, masks, cuboids, key‑points, etc.).
• Integrating dozens of pretrained models for tasks such as detection, segmentation, OCR, pose estimation, depth, VQA, and even multimodal grounding.
• Supporting both local inference (ONNX, TensorRT, OpenCV DNN, vLLM, SGLang) and remote inference via the companion X‑AnyLabeling‑Server.
• Exporting to all common dataset formats (COCO, VOC, YOLO, MOT, DOTA, PPOCR, ShareGPT, …).

Key capabilities

Category	What you can do
Data types	Images, video, plain text, PDFs, 3D point clouds, multimodal bundles
Annotation tasks	Classification, object detection (HBB/OBB), instance/semantic segmentation, pose, face, tracking, OCR/KIE, lane marking, depth, image captioning, VQA, tagging, grounding, counting
Tools	Polygon, rectangle, rotated box, cuboid, circle, polyline, point, mask, magic‑wand (region‑grow), interactive SAM, video‑object‑segmentation, etc.
Model library	Over 40 ready‑to‑use models (YOLOv5‑v12, YOLOX, YOLO‑NAS, SAM 1/2/3, Grounding‑DINO, Florence‑2, Qwen‑VL, etc.)
Inference back‑ends	ONNX Runtime, TensorRT, OpenCV DNN, vLLM, SGLang, plus custom remote servers
Formats	Import/export COCO, VOC, YOLO, DOTA, MOT, MASK, PPOCR, MMGD, VLM‑R1, ShareGPT, …
Platforms	Windows, Linux, macOS; UI in English, Chinese, Japanese, Korean

Typical workflow

1. Load your raw data (folder of images, video file, point‑cloud file, etc.).
2. Select a model from the built‑in library (e.g., YOLO‑v8 for detection, SAM‑2 for segmentation).
3. Run inference – the app sends the data to the chosen engine (local GPU/CPU or remote server) and receives predictions.
4. Edit the auto‑labels using the drawing tools; you can refine masks, adjust boxes, add key‑points, or correct OCR text.
5. Export the final annotations in the format required by your training pipeline.

Getting started

• Install via pip: pip install x-anylabeling-cvhub (requires Python 3.11+).
• Or download the pre‑built desktop binary from the GitHub releases page.
• Run xanylabeling to launch the GUI, then follow the Installation & Quickstart guide in the docs.

Extensibility – You can plug in your own model by providing a Python wrapper that conforms to the X‑AnyLabeling inference API, or point the app at a remote X‑AnyLabeling‑Server that hosts custom models and GPU resources.

Community & support – Open‑source under GPL‑v3, with a contribution guide, issue tracker, and a sponsorship page for funding continued development.

---

TL;DR – X‑AnyLabeling is a feature‑rich, multi‑modal annotation workstation that couples a huge model zoo with an easy‑to‑use GUI, making it fast to generate and polish training data for a wide range of computer‑vision tasks.