benjaminplanche
2018-06-13 22:20:43
You could use torch.nn.AvgPool1d (or torch.nn.AvgPool2d, torch.nn.AvgPool3d) which are performing mean pooling - proportional to sum pooling. If you really want the summed values, you could multiply the averaged output by the pooling surface.
热门github
6
Do you mean, first I perform Avgpool2d and than multiply the resulting tensor by 4 if my kernel size is 4 ?
Yes, average pooling first, then multiply back the results by
4 * 4 = 16ifkernel_size=4(as it means the average is computed over a region of4 * 4 = 16elements).