Dear all,
I would like to freeze not just the last fully connected layer of EfficientNet-b0 but also some of the previous block to apply transfer learning to a fairly different domain.
This is something I have been able to achieve with ResNet18:
cntr=0
for name, child in model.named_children():
cntr+=1
if cntr < lt:
print(f'{name} not trainable')
for param in child.parameters():
param.requires_grad = False
else:
print(f'{name} trainable')
But found complex with EfficientNet!