Why concatenating a matrix of ones?

[pmorerio]

Hi and first of all thanks for the great code!

Can I ask why you stack a matrix of ones in the penultimate channel here?

Diverse-Structure-Inpainting/net/structure_generator.py

Line 30 in 467b7a9

x = tf.concat([x, ones_x, ones_x*mask], 3)

Wouldn't it make more sense to have x = tf.concat([x, ones_x*(1-mask), ones_x*mask], 3) ?
(i.e. you concat the mask as its one-hot encoding).

Best,
P.

[USTC-JialunPeng]

It's a good question.
It originated from the implementation of PixelCNN++, a widely used PixelCNN.
https://github.com/openai/pixel-cnn/blob/bbc15688dd37934a12c2759cf2b34975e15901d9/pixel_cnn_pp/model.py#L36
However, I think it may have little impact on the performance. You can try to remove it and re-train the model.

Best wishes,
Jialun

[pmorerio]

Thanks for your answer!