SO101 rack-handle segmentation
Gripper-camera segmentation model trained from 177 manually reviewed positive masks and 72 reviewed negatives.
Held-out test metrics
- Mask mAP50-95: 0.3145
- Mask mAP50: 0.8930
- Mask precision: 0.9274
- Mask recall: 0.8523
The split is grouped by source episode. The 20-image test split was not used for optimization or early stopping.
