OICR-PyTorch
OICR-PyTorch copied to clipboard
About the performance.
It is weird that the mAP I got with your codes is about 37~38. I use the default setting. I also can not reproduce the results of https://github.com/luiszeni/Boosted-OICR.
PS: Pytorch ==1.3 cuda==10.1 rtx2080ti
My Pytorch version is 1.2.
I think maybe CUDA and GPU affect the performance, sometimes the old version of the hardware can be more stable.
You can test with my model, I have updated it in README