Question about image preprocessing methods in SAM1 and SAM2

[qirui-chen]

Hello authors, thank you for your excellent work. I would like to ask why, in SAM1, the image is first resized by its long side to 1024, and then padded for the short side, whereas in SAM2, the image is directly resized to (1024, 1024) using PIL.Image's resize method. Wouldn't this cause distortion of the aspect ratio?

Additionally, which resolution and which pixels are the loss applied to during training?

Thank you!