LEDNet
LEDNet copied to clipboard
Incorrect order for APN
Thank you for your great work!
But I may find a mistake for the APN implementation. According to Pyramid Attention Networkfor Semantic Segmentation and the official implementation, the kernel sizes for convolutions with stride=2 should be 7x7, 5x5, 3x3 in order, while your implementation is 3x3, 5x5, 7x7.
Would you mind to provide an alternative pretrained model?