Hi, could you please help me understand the difference between version b, c, d, e, s of ResNetV1b in GluonCV? And why their dilated version is used as backbone in Segmentation models?
This papers goes over the difference between the different letter variants of ResNet
b, c, d are explained there.
V1e is based on v1d, the beginning of the channel changed from 32-32-64 to 64-64-128
v1s is based on v1c, the beginning of the channel changed from 32-32-64 to 64-64 -128
The dilation is used to increase the receptive field for the output of the layer.
Also see this answer here on the discuss gluon forum: https://discuss.gluon.ai/t/topic/9656
Thanks a lot! I was clueless