I found when import face_alignment package, it set torch.set_grad_enabled(False), which makes torch close autograd. I was planning to use this package to extract landmarks for my training task and it created some trouble because autograd was closed. I spent some time to finally figure this out.
I think it will be a good feature if autograd is re-enabled after loading model.
I found when import
face_alignment
package, it settorch.set_grad_enabled(False)
, which makes torch close autograd. I was planning to use this package to extract landmarks for my training task and it created some trouble because autograd was closed. I spent some time to finally figure this out. I think it will be a good feature if autograd is re-enabled after loading model.