Hello,
I made faster rcnn using c++ frontend based on maskrcnn-benchmark structure.
https://github.com/lsrock1/maskrcnn_benchmark.cpp
Using trained weight in python(maskrcnn-benchmark) i successfully reproduced the original performance res50 and res101
The Parallel module of my implementation is not perfect. So, i couldn’t test training yet.
But this repo includes a script about python weight to jit and jit to c++ tensor and almost every functions in original repo(data loading, transform, weight loading etc.) with same structure.
I think this implementation could be interesting.
Thanks!