PyTorch 1.4 / torchvision 0.5 vs Detectron

Hi guys,

I just noticed that with the latest PyTorch 1.4 release: torchvision 0.5 provides better deployment support in terms of exporting to ONNX and quantisation. What’s the timeline for the same features on detectron2?

Also, there seems to be significant overlap between torchvision and detectron. What’s the vision for the two? Do you recommend one over the other for prototyping and deployment?