Modifying Mask R_CNN Architecture

I am working on a research project and as part of it, I need to modify the Mask R_CNN architecture to have multiple inputs and outputs. I want to choose a base code for this, but various repos are available, including detecron2 and Facebook Mask_R_CNN. I found detecron2 complicated to start with. Do you suggest using PyTorch implementation of Mask R CNN?

Hi Samin!

I have used pytorch’s Mask R-CNN. It works and seems sensible and
solid, but I don’t have any useful comparisons with other implementations.

One advantage of using the pytorch version is that you might be a little
more likely to get answers about it on this forum, rather than about other


K. Frank

1 Like