Hi-SAM
Hi-SAM copied to clipboard
train
Can the Hi-SAM be trained to segment the watermark, such as some lines?
It seems feasible as long as the mask of watermask can be achieved.