Why Resnet first conv layer has kernel_size=3 and not =7, like the original paper?

[acnazarejr]

face_recognition.pytorch/model/backbones/resnet.py

Line 24 in b5c4be9

self.conv1 = nn.Conv2d(3, 64, kernel_size=3, stride=1, padding=1,

[grib0ed0v]

@acnazarejr actually, 7x7 convolution is much heavier than 3x3, that's why we decided to choose 3x3. Since we don't use pre-trained weights from ImageNet, that's should be OK.