search

aai-institute / ProtoTree

Fork of ProtoTrees: Neural Prototype Trees for Interpretable Fine-grained Image Recognition, published at CVPR2021
Other
3 stars 0 forks source link

ProtoTree for multi-label classification #59

Open sissaNassir opened 6 months ago

sissaNassir commented 6 months ago

Very interesting work! I was wondering if ProtoTree can be applied in multi label classification setting, say in the domain of medical image analysis. If not, do you know what are the short comings that stop it from being applied there? Thank you!

MischaPanch commented 5 months ago

Hi. I agree, these models are interesting :)

Development here stalled but may be continued soon. In that case we will probably move to another repo, doesn't make much sense to continue from the fork.

In principle, prototype based models can be applied for multi-label classification just as other models. This project is not currently in a state that would allow such a feature extension though, we'd first need to turn in into a proper library.

If you're interested, I can keep you updated on further developments

sissaNassir commented 5 months ago

Thanks for the response! I discovered your work through Nauta. I'm just starting approaching to these models. If I can ask, why did you remove last convex optimization layer and the pruning in protopnet?

MischaPanch commented 5 months ago

These elements of protopnet add a lot of complexity to the training, and it doesn't seem like they actually bring anything in terms of performance. So we removed them, which also allowed us to largely unify the logic and the code of protopnet and prototree