Transfer learning: feature extraction

[Salvatore-tech]

Describe the bug
Good evening, i was wondering if some high level functions are provided to perform feature extraction on a model as RAFT.
I'd like to load the pretrained model on Things3d+Chairs and update only the final layer of it.

[Zachary-66]

I don't clearly understand what you mean, there may be several understandings:

1. The final layer is a singel layer, such as a Conv layer. For now, we don't have such high level function.
2. You want to update the parameters of a complete submodule and freeze the other submodule. For example, when training FlowNet2CS, which is composed of flownetC and flownetS1, we only want to update flownetS1. By referring to the config, just set the freeze_net of flownetC to True.