Syncing with the implementation of official repo
seermer opened this issue · 1 comments
seermer commented
The official repo made some changes that are different from the paper.
For example, the paper claims that v2-s uses 272 channels in the last stage, but they changed to 256 in their code, and the author claimed that the code will be the source of truth. Is this repo planning on syncing with the official code?
Thanks
launchauto commented
Yes, it is 256
efficientnetv2.pytorch/effnetv2.py
Line 179 in 775326e
the c column indicates the output of efficientnetv2-s backbone (before the last layer conv-avgpool-linear) is 256.
Or you can put the code in your program and see the output tensor size.