X-AnyLabeling

What the application covers

X AnyLabeling is a lightweight, efficient, and unified cross platform desktop application for AI assisted annotation of text, image, video, and multimodal data. It combines built in tools, automated labeling workflows, deep learning models, and flexible import and export of several formats. The project runs on Windows, Linux, and macOS, and its interface is available in English, Simplified Chinese, Japanese, and Korean. The README's badge row states a GPL version 3 license, a Python 3.11 or newer requirement, and support across Windows, Linux, and macOS, and it links a ModelScope collection for model downloads. The tool covers a wide range of tasks. These include image classification, object detection, instance segmentation, pose estimation, oriented object detection, multi object tracking, optical character recognition, lane annotation, image captioning, visual question answering, and document parsing. The shape tools provided are polygons, rectangles, cuboids, rotated boxes, quadrilaterals, circles, lines, polylines, points, masks, and task specific tools for text detection, text recognition, and key information extraction. For inference it supports both local and remote execution through engines such as ONNX Runtime, TensorRT, OpenCV DNN, vLLM, and SGLang. Remote inference is handled by a companion project called X AnyLabeling Server, which is a lightweight backend for connecting custom models and compute. Recent changes include image tagging added on 2026 08 19, D FINE seg on 2026 08 12, RT DETRv2 OBB and a magic wand tool on 2026 08 08, and the 4.0.0 release on 2026 08 05. The README also notes that the project has been actively maintained and that community contributions are welcome through its contributing guide and contributor license agreement.

Models and data formats

The model library spans many task categories and names specific supported models. Image classification includes YOLOv5 Cls, YOLOv8 Cls, YOLO11 Cls, InternImage, and PULC. Object detection includes YOLOv5 through YOLO12, YOLOX, YOLO NAS, D FINE, DAMO YOLO, Gold YOLO, RT DETR, RF DETR, and DEIMv2. Instance segmentation includes YOLOv5 Seg through YOLO26 Seg, Hyper YOLO Seg, RF DETR Seg, and D FINE seg. Pose estimation includes YOLOv8 Pose, YOLO11 Pose, YOLO26 Pose, DWPose, and RTMO. Rotated object detection includes YOLOv5 Obb through YOLO26 Obb and RT DETRv2 OBB. Other categories include depth estimation with Depth Anything, segment anything with SAM 1 through 3, image matting with RMBG, tagging with RAM and RAM++, OCR with PP OCRv4 through v6, layout analysis with PP DocLayoutV3, document parsing with PaddleOCR VL, vision language models such as Qwen3 VL, Gemini, ChatGPT, and GLM, lane detection with CLRNet, and grounding with Grounding DINO, YOLO World, YOLOE, SAM 3, and LocateAnything. For import and export the tool supports COCO, VOC, YOLO, DOTA, MOT, MASK, PPOCR, MMGD, VLM R1, and ShareGPT. The README notes the table is not the whole list and points to a model zoo document for the rest, and that image tagging in the recent changes supports tag creation, editing, reordering, and batch deletion.

Running and extending the tool

The README points to an installation and quickstart guide, a usage guide, a command line interface document, a custom model guide, and guides for the chatbot, visual question answering, image classifier, video classifier, and document parsing with intelligent text recognition. The examples folder walks through classification at image level and shape level, detection for horizontal and rotated boxes, several kinds of segmentation, description tasks such as tagging and captioning, face and pose and depth estimation, optical character recognition with text recognition and key information extraction, multiple object tracking by several methods, interactive video object segmentation with SAM2 and SAM3, matting, vision language examples with Rex Omni and Florence 2, counting with GeCo and GeCo2, and grounding with YOLOE, SAM 3, and LocateAnything. Training examples cover Ultralytics. The license is the GNU General Public License version 3.0, which permits use, modification, and redistribution including for commercial purposes as long as you comply with the terms. The project credits several earlier annotation tools whose work informed it, including AnyLabeling, LabelMe, LabelImg, roLabelImg, PPOCRLabel, and CVAT. If you use the software in research, the README supplies a citation block that names the author as Wei Wang and the publisher as CVHub on GitHub.

Editorial conclusion

X AnyLabeling is a cross platform annotation desktop application licensed under GPL version 3.0. It supports Windows, Linux, and macOS, imports and exports formats such as COCO, VOC, YOLO, and DOTA, and bundles a large model library for assisted labeling.