search

cvlab-columbia / pix2gestalt

Code for the paper "pix2gestalt: Amodal Segmentation by Synthesizing Wholes" (CVPR 2024)
https://gestalt.cs.columbia.edu/
Other
119 stars 8 forks source link

Raw images before pasting foreground objects #13

Closed zhyever closed 3 weeks ago

zhyever commented 3 weeks ago

Thanks for your amazing work! I was wondering if it would be possible that you also release the raw images before pasting the foreground objects. Or would it be possible that we can use some strategies to find that corresponding raw images from the current SAM-1B dataset?

Also, could you please share the code for foreground check? It could be very helpful. Thanks.

zhyever commented 3 weeks ago

The id is exactly the same as the raw sam-1b dataset

egeozguroglu commented 3 weeks ago

Thanks for your interest in our paper. Correct, you can use the SA-1B IDs directly.