Enes Erdogan

Images were read by using np.fromfile function, I guess code was designed to read raw images. To debug that you can change that part with the cv2.imread

Tihs comment solved my problem: [click](https://github.com/NVlabs/PoseCNN-PyTorch/issues/24#issuecomment-1085552217)

I only installed these and it works ``` pip install fvcore conda install pytorch torchvision torchaudio cudatoolkit=10.2 -c pytorch # Currently it is pytorch 1.12 pip install opencv-python==4.6.0.66 ```

@vivekh2000 I did not check the paper, but I think @jacobgil also implemented the `discard_ratio` which may not be available in the paper because this obviously breaks the normalization of...