search

DingXiaoH / RepLKNet-pytorch

Scaling Up Your Kernels to 31x31: Revisiting Large Kernel Design in CNNs (CVPR 2022)
MIT License
860 stars 86 forks source link

Using DW 3x3 in stem block #49

Open quannguyen268 opened 1 year ago

quannguyen268 commented 1 year ago

Thank you for such a great paper, however I have some questions. In section 4.1, you stated in the paper that : "we arrange a DW 3×3 layer to capture low-level patterns". Can you explain more about :

  1. How DW 3x3 layer can capture low-level patterns ?
  2. Does Conv 3x3 has that attribute ?
  3. Why don't you use 1 conv 3x3 instead of dw 3x3 and conv 1x1 ?