The situation is the following. A U-Net with a resnet34 backbone is trained. Is there a way to reuse the fine-tuned backbone part of the U-Net to use it to perform more transfer-learning for other tasks? If so, how do I get this “sub-network” from the U-Net?
You can do it. You can simply choose a part of your entire architecture. The code would look like this.
list(learn.model.modules())[0]
4 Likes