Will you consider video input

[AmmonZ]

It seems that for some cases like suitcase (like randomly pick one image from amazon), it cannot generate the prismatic part in the final urdf (which is actually not sim-ready). Will it help if this method can use human video input as done by previous sota?

[ziangcao0312]

Using videos as input may improve performance, but videos are harder to obtain than single images. Considering data availability, we use a single image as the input. If you want to further improve the results, once we release the training code, you can consider training a model that takes video as input.