pycococreator

Why we included this project

Training instance or panoptic segmentation models usually means feeding them ground-truth annotations in the COCO JSON format. This small Python library converts binary masks, the kind you get from a labeling tool or synthetic renders, into the uncompressed RLE and polygon structures COCO expects, so your data is ready for standard training pipelines. It's a focused utility rather than a full labeling suite, so it fits best when you already have masks and just need reliable conversion into COCO-compatible annotation files. Teams building custom datasets for Mask R-CNN, Detectron, or similar detectors will find it a straightforward way to get from mask files to the annotation files those frameworks load.