The code is found on line 475 of EfficientNet-PyTorch/efficientnet_pytorch/utils.py. According to formula 3 of the original paper, the image size of efficientnet-b7 is 600. r=γ^Φ=1.15^Φ=600/224=2.6786, so Φ=log(2.6786)/log(1.15)=7.04. Therefore, depth=d=α^Φ=1.2^7.04=3.6094≠3.1
I have browsed several repos about EfficientNet, which most of them define depth of efficientnet-b7 as 3.1, so I'm not sure if this derivation is correct. If I'm wrong, what's the correct mathematical derivation?
The code is found on line 475 of EfficientNet-PyTorch/efficientnet_pytorch/utils.py. According to formula 3 of the original paper, the image size of efficientnet-b7 is 600. r=γ^Φ=1.15^Φ=600/224=2.6786, so Φ=log(2.6786)/log(1.15)=7.04. Therefore, depth=d=α^Φ=1.2^7.04=3.6094≠3.1
I have browsed several repos about EfficientNet, which most of them define depth of efficientnet-b7 as 3.1, so I'm not sure if this derivation is correct. If I'm wrong, what's the correct mathematical derivation?