Creating a chainercv like interface for pytorch

Hi. I was wondering if there is any interest in creating a chainercv( like interface for pytorch using fastai. I would say classification and regression parts are already there however a lot more could be added. For example, support for recent object detectors/segmentation/instance segmentation/gan models etc. I would say most of the code is already there in the notebooks, its just a matter of easier use of the interface. Having a model zoo on top of that would be perfect.

Interested to know your thoughts.

Yes that’s the next thing we want to work on! :slight_smile: We’ve got most of the pieces in place. If you’re interested in contributing, a good first step would be implementations of the main object detector architectures.

Segmentation already works BTW.

1 Like

Amazing. I have a retinanet and ssd code in progress, though I am not reaching the required performance so still in the process of debugging. I have my hands tied with some other work, but I should be able to get something running in next 2 weeks. Will keep you updated.