Suppose, I have a PyTorch tensor a of shape ba,c,h,w and wish to zero out some rows of a with indices given in another tensor b of shape ba,2 of dtype=torch.int16 and b[batch, 0] <= b[batch, 1]. The for loop way of doing is: for batch in range(ba): a[batch,:,0:b[batch,0],:] = 0 # stmnt 1 a[batch,:,b[batch,1]:,:] = 0 # stmnt 2 Is there a faster PyTorch way to do this? Specifically, firstly by combining stmnt1 and stmnt2 into a single line to tell PyTorch make all rows of a except a[batch,:,b[batch,0]:b[batch,1],:] zero. And secondly, if this can be done without needing to iterate over each batch using a for loop. Continue reading...
