Im2Contact: Vision-Based Contact Localization Without Touch or Force Sensing

im2contact on Out of Distribution Objects
Note: Videos below are with model trained on 383.3 hours of simulated time instead of the 187.5 hours used for ablation comparisons

Mock mixing with a non-convex spatula and bowl (Fig 1)

In-hand reorientation of a non-convex mug followed by hanging on a rack (Fig 8 Right)

Interaction between two moderately deformable plushies (Fig 8 Left)