21 object categories from the YCB-Video dataset and synthetic sequences designed for 6D pose tracking. The data provides RGB-D frames, object segmentation masks, and ground truth SE(3) transformations representing relative motion between consecutive frames.
Use Cases
- Train a deep neural network to predict relative pose changes using the image_residual and ground_truth_pose features.
- Benchmark 6D pose tracking algorithms by calculating the ADD (Average Distance) metric against the provided SE(3) labels.
- Implement object-centric filtering by applying the provided segmentation masks to RGB-D input streams.
Strengths
- Includes 21 distinct object classes from the YCB-Video benchmark for 6D pose estimation.
- Provides ground truth SE(3) transformations consisting of 3D rotation and translation vectors for sequential frames.
- Contains synthetic image pairs used to generate residual maps between rendered and observed object poses.
- Features RGB-D data and binary segmentation masks for each tracked object.