Use SAM_Perceptual(train=True, weights=[0,1,1,0]) as training loss instead of IMPACTReg 9833ede verified VBoussot commited on 4 days ago