ERFNet-SAD

[voldemortX]

When you do ERFNet-SAD, can you tell me exactly which layer learns from which?
I've tried to implement ERFNet-SAD but can't find a way at it, at the moment.

cc @cardwing

I'm might have talked about doing ERFNet-SAD in your other repo, but I can't remember where... So I opened a new one here.

[cardwing]

Can you refer to the issues in that repo? It's hard for me to give you the answer as many old codes have been cleaned several days ago.

[voldemortX]

Can you refer to the issues in that repo? It's hard for me to give you the answer as many old codes have been cleaned several days ago.

I mean I mentioned implementing ERFNet-SAD in TuSimple, but I failed as of yet.

[voldemortX]

@cardwing Found it (finally) here. It was a non-related old and closed issue, I think we should move the discussions here since it is in this paper ERFNet-SAD is introduced.

You see I'm working on a unified framework in pytorch that do not rely on lua or matlab (we have already adapted ERFNet from your older pytorch version codes and achieved slightly better performance):
https://github.com/voldemortX/pytorch-auto-drive

I'd love to have SAD added in the best backbone in the framework (ERFNet). But I want to do it right so can I have some more references, such as where the SAD loss is added?