transmuteAI/trailmet

MobileNetv2 skip connections differing from torchvision implementation

Opened this issue · 0 comments

sydarb commented

There shouldn't be any skip connections for inverted residual blocks having input channel != output channel.
ref : https://github.com/pytorch/vision/blob/main/torchvision/models/mobilenetv2.py#L32

if stride == 1 and in_planes != out_planes: