Open YaphetS7 opened 1 year ago
kernel_filter was computed two times: first for out, second for label. There was a bottleneck in first branch - the reason is .cuda() method. Now there are no bottlenecks in the code.
kernel_filter was computed two times: first for out, second for label. There was a bottleneck in first branch - the reason is .cuda() method. Now there are no bottlenecks in the code.