Minimal working example produces red tinted images on forward inference about pynet-pytorch HOT 2 CLOSED

So, I have made some progress on this issue.

It seems like it is related with the decomposition of the Bayer filter. I tried a different decomposition and it gets closer to the real RGB picture, but now the colours are much warmer than in the original picture.

def extract_bayer_channels(raw):
    ch_R = raw[0::2, 1::2]
    ch_Gr  = raw[0::2, 0::2]
    ch_B = raw[1::2, 0::2]
    ch_Gb  = raw[1::2, 1::2]
    bayer = np.dstack((ch_B, ch_Gb, ch_R, ch_Gr))
    return bayer

For reference, this is the original picture:

from pynet-pytorch.

I have just realized that I was using a different dataset, S7 ISP. That is why I got different odd image colours.

I have also realized that this is caused by different camera white balance settings. If I use the daylight white balance to convert the RAW image from S7 ISP instead of the camera, I also get RGB images with warmer tones.

from pynet-pytorch.