Hello all.
I’d like to implement a function:
-
argument:
– x (torch.Tensor): x.shape == (B, C, H, W)
– recipe (maybe python:dict): ex) {1: (0, 50), 2: (50, 75), 3: (75, 100)} -
returns:
– mask (torch.Tensor): mask.shape == (B, C, H, W) -
how it works:
In below figure, white matrix represents only one channel of given x.
(i.e., x[sample_batch_idx][sample_channel_idx])
also arbitrary recipe is given by proportional.(not threshold)
the function returns mask represented by colored matrix.
I’d like to do this process in channel-wisely.
Currently, I implemented a test function using for loop, but It shows horrible performance.
Is there any efficient way to implement this function?
I think the (automatic) broadcasting will be the key idea instead of for loop, but I don’t know how to apply this.
Any suggestions will be welcome for me.