yolov7
yolov7 copied to clipboard
model.fuse() vs re-parameterization ?
What's the difference model.fuse() and reparameterization of below link?
https://github.com/WongKinYiu/yolov7/blob/main/tools/reparameterization.ipynb
I have the same question.
As far as I can see, fuse
methods are present, but they are not used anywhere, except here
For yolov7
model in provided notebook only Detect
head is fused, but 'Conv' with BatchNormalization
are don't.
It seems, model.fuse()
is not implemented alternative to manual reparameterization
Any conclusion on this? From what I understand the reparameterisation given in the repameterization.ipynb is like copying the weights of ema. And its different from the one mentioned in the paper.