Have you evaluated the SAM-6D PEM module outside of the zero shot setting?
One obvious path would be to train on the BOP datasets either instead of the large-scale training, or as further training on top of those weights.
Another less obvious path would be to use the training data to construct Fo/Po instead of the renders.
Your thoughts appreciated. Feel free to reach out to aleksey@apera.ai
Have you evaluated the SAM-6D PEM module outside of the zero shot setting?
One obvious path would be to train on the BOP datasets either instead of the large-scale training, or as further training on top of those weights.
Another less obvious path would be to use the training data to construct Fo/Po instead of the renders.
Your thoughts appreciated. Feel free to reach out to aleksey@apera.ai