COGAR Simulated Robotic Scene Segmentation Datasets

This dataset release includes two synthetic robotic-scene segmentation datasets created for the project "Foundation Model Segmentation for Robotic Scenes" The publication comprises: 1. Isaac official Unitree G1 synthetic dataset: 1000 RGB images produced in Isaac Sim utilizing the official Unitree G1 asset, accompanied by segmentation masks, metadata, and COCO-format annotations. 2. BlenderProc COGAR-SimRobotics-1000 dataset: 1000 RGB images produced using BlenderProc, featuring reflective metal, transparent glass, partial occlusions, small components, and dynamic scene complexities, accompanied by metadata and COCO-format annotations. The datasets were utilized to evaluate SAM, SAM2, FastSAM, MobileSAM, EfficientSAM, YOLOv8-seg, Mask R-CNN, and DeepLabV3+ for robotic scene segmentation. SHA256 checksum files and release manifests are provided for reproducibility. OCID is not included in this release; users must acquire OCID from its original source.