Channels is not consistent when it is computed in shortcut layer in resnet50 cfg

[lixiaohui2020]

In resnet50 cfg ` [net] batch=1 subdivisions=1

height=256 width=256 max_crop=448 channels=3 momentum=0.9 decay=0.0005

....................

[convolutional] batch_normalize=1 filters=64 size=7 stride=2 pad=1 activation=leaky

[maxpool] size=2 stride=2

[convolutional] batch_normalize=1 filters=64 size=1 stride=1 pad=1 activation=leaky

[convolutional] batch_normalize=1 filters=64 size=3 stride=1 pad=1 activation=leaky

[convolutional] batch_normalize=1 filters=256 size=1 stride=1 pad=1 activation=linear

[shortcut] from=-4 activation=leaky `

After maxpool layer, we get 64 X 64 X 64 Before shortcut layer, we get 64 X 64 X 256 I don't understand how 64 X 64 X 64 and 64 X 64 X 256 are computed in shortcut layer.

I check KaimingHe

https://github.com/KaimingHe/deep-residual-networks/blob/master/prototxt/ResNet-50-deploy.prototxt

There is a filter 1 X 1 X 256 to convert 64 X 64 X 64 to 64 X 64 X 256. So I think the resnet50 cfg should add a conv layer, such as ` [net] batch=1 subdivisions=1

height=256 width=256 max_crop=448 channels=3 momentum=0.9 decay=0.0005

....................

[convolutional] batch_normalize=1 filters=64 size=7 stride=2 pad=1 activation=leaky

[maxpool] size=2 stride=2

[convolutional] batch_normalize=1 filters=256 size=1 stride=1 pad=0 activation=linear

[convolutional] batch_normalize=1 filters=64 size=1 stride=1 pad=1 activation=leaky

[convolutional] batch_normalize=1 filters=64 size=3 stride=1 pad=1 activation=leaky

[convolutional] batch_normalize=1 filters=256 size=1 stride=1 pad=1 activation=linear

[shortcut] from=-4 activation=leaky `

I want to know whether my suggestion is right. Sincerely hope for your reply.

[liuyang-zhx]

Hello, I also encountered this problem, have you solved it? Thank you

[qianhang1994]

Hello, met same question, have you solved it? Thank you