SynTable-Sim
Sample Synthetic Dataset Generated using SynTable Pipeline
About
To demonstrate the capabilities of SynTable, we generated a sample synthetic dataset, SynTable-Sim, using our pipeline, for training and evaluating UOAIS models. Note that users can also generate other custom datasets that meet the specific requirements of their application using SynTable.
Figures
Total Images: 50,000 RGBD Images (45,000 Training, 5,000 Validation)
Tabletop Scenes: 1000 Scenes, 50 different viewpoints for each scene.
Training Dataset Object Composition: 1075 object models from the Google Scanned Objects dataset and the Benchmark for 6D Object Pose Estimation (BOP).
Validation Dataset Object Composition: 78 novel objects from the YCB dataset.
Table Object Composition: Sampled 10 Omniverse Nucleus table assets to provide randomization for each scene.
SynTable Training & Validation Dataset
Train Dataset
Validation Dataset
Dataset Statistics
Histogram for number of connected components for UOAIS-Sim tabletop, SynTable-Sim and OSD-Amodal datasets
Histogram for number of regions per connected component (connected component size) for UOAIS-Sim tabletop, SynTable-Sim and OSD-Amodal datasets
Histogram of occlusion rate for UOAIS-Sim tabletop, SynTable-Sim and OSD-Amodal datasets