Join GitHub today
GitHub is home to over 40 million developers working together to host and review code, manage projects, and build software together.Sign up
Add PyTorch backends: first pass #821
This PR adds chip classification and semantic segmentation backends based on PyTorch and fastai, along with a lot of infrastructure to simultaneously support the existing TF backends. Initially, I tried to create a single Docker image containing everything, but that approach was problematic. The existing version of Tensorflow in RV wasn't compatible with the latest PyTorch/Fastai because of a conflict in the versions of numpy and cuda. I could have tried to upgrade the three backends to use the latest version of Tensorflow, but I thought that was risky and not worth the effort if we're sunsetting the TF backends. So, I decided to just have separate images, which was quite a bit of work in and of itself.
A subsequent PR will add the following: