--- license: cc-by-nc-4.0 task_categories: - image-segmentation - object-detection - robotics tags: - synthetic-data - 6d-pose - bin-picking - blender-cycles - transparent-objects - depth-estimation pretty_name: TransReflect-500 Robotics Perception Dataset (100-Scene Sample) --- # TransReflect-500: Synthetic 6D Pose & Segmentation for Transparent/Reflective Robotics 100-scene public preview slice from our domain-randomized **Blender Cycles (OptiX)** synthetic data pipeline for robotic bin-picking, lab automation, and 6D pose estimation on hard-to-sense materials (**clear glass/acrylic, glossy plastics, and reflective metals**). ![4-Scene QA Grid: RGB, Millimeter Depth, Instance Masks, and 6D Pose Axes](qa_grid.png) ## Dataset Structure ```text TransReflect-500-Robotics/ ├── qa_grid.png # 4-scene visual QA sheet (RGB, depth, mask, 6D pose) ├── dataset_info.json # Global camera intrinsics & generation metadata ├── rgb/ # 8-bit photorealistic RGB renders (0000.png – 0099.png) ├── depth/ # 16-bit depth maps in millimeters (0000.png – 0099.png) ├── masks/ # Pixel-exact integer instance segmentation masks (0000.png – 0099.png) └── labels/ # Per-frame JSON annotations (0000.json – 0099.json)