Skip to content
This repository has been archived by the owner on Sep 27, 2024. It is now read-only.

Why Resnet first conv layer has kernel_size=3 and not =7, like the original paper? #2

Open
acnazarejr opened this issue Apr 8, 2019 · 1 comment
Labels
question Further information is requested

Comments

@acnazarejr
Copy link

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

@grib0ed0v
Copy link
Owner

@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.

@grib0ed0v grib0ed0v added the question Further information is requested label Apr 8, 2019
Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.
Labels
question Further information is requested
Projects
None yet
Development

No branches or pull requests

2 participants