imatge-upc / salgan

SalGAN: Visual Saliency Prediction with Generative Adversarial Networks
https://imatge-upc.github.io/salgan
MIT License
367 stars 106 forks source link

Color channels on input images are shuffled #33

Open filipux opened 6 years ago

filipux commented 6 years ago

I downloaded your pretrained models and got your code working on my macbook with great results. However, when I did some experimentation I noticed the colors of the image_stimuli in the predict function where wrong. I then found that when the images are being loaded on line 17 in 03-predict.py you seem to shuffle the color channels:

cv2.cvtColor( ... , cv2.COLOR_BGR2RGB)

What is the reason for doing this? Have your pretrained models been trained using shuffled image channels too?