I open-sourced ParkerLabel, a desktop tool for object detection and instance segmentation annotation. It runs locally on the CPU. No GPU or image upload is needed.
In AI mode, positive and negative clicks guide MobileSAM to generate an object mask. You can correct it manually, assign a category, and save the annotation. The interface has separate image, mask, and overlay views.
The inference path uses ONNX Runtime. The image embedding is saved beside the image and reused. Each instance keeps its own mask, category, bounding box, and area in a per-image JSON file. Masks use uncompressed COCO RLE; these files are not a complete COCO dataset export.
There are portable packages for macOS arm64 and Windows x64. Extract the folder to a writable location and run the app; no Python environment is needed. The UI supports nine languages, including English and Chinese.
If you work with image annotations, give it a try. Bug reports and workflow feedback are welcome.
Top comments (0)