Although original paper https://arxiv.org/pdf/2104.00298.pdf uses Squeeze and Excite block in FusedMbConv (see Figure 2 in the paper), torchvision’s implementation does not use them. What is the reason? And why is that not documented?
Although original paper https://arxiv.org/pdf/2104.00298.pdf uses Squeeze and Excite block in FusedMbConv (see Figure 2 in the paper), torchvision’s implementation does not use them. What is the reason? And why is that not documented?