Bharat Singh

you need to change the evaluation code (lib/pycocotools/cocoeval.py, accumulate and summarize) and save the stats.

I didn't do it for pascal, so i don't know how to do it, but it should be possible to do.

Use this in module.py. Evaluating metric in each iteration is expensive. Also, mxnet does asynchronous calls to forward and backward, so the prefetcher can run in background. Anchor generation for...

I think you can get around 10 with 8 gpus (titan xp/1080Ti), I am not sure whats wrong here.

thanks, that cut the latency down from 5 seconds to around 1 second. Is there a way to bring it down to 0.2 seconds or something? I switched to streaming...

https://github.com/bharatsingh430/py-R-FCN-multiGPU You can use this branch

any update on this?

The updated msra repo has the code for multi scale testing

Thanks for the code!. Using pytorch ops may not be efficient for the GPU version for soft-nms. It would probably need a custom CUDA implementation where each loop can be...