Dear Dr. Foster,
I am studying Chapter 10, “Advanced GANs,” specifically the ProGAN section. While reviewing Figure 10-4, which shows the ProGAN discriminator training process when expanding from 4 × 4 images to 8 × 8 images, I noticed a possible labeling issue.
In the transition and stabilization parts of the discriminator diagram, the blue block is labeled “New toRGB.” However, since this is the discriminator, the layer appears to be converting an RGB image into feature maps. Therefore, it seems that the label should be “New fromRGB” rather than “New toRGB.”
My understanding is that:
- In the generator, toRGB converts feature maps into an RGB image.
- In the discriminator, fromRGB converts an RGB image into feature maps.
Because Figure 10-4 describes the discriminator, I believe the blue block should be labeled “New fromRGB.”
Thank you for your excellent book. I am learning a lot from it, and I wanted to share this possible correction respectfully.
Sincerely,
Basil Aljarih
Dear Dr. Foster,
I am studying Chapter 10, “Advanced GANs,” specifically the ProGAN section. While reviewing Figure 10-4, which shows the ProGAN discriminator training process when expanding from 4 × 4 images to 8 × 8 images, I noticed a possible labeling issue.
In the transition and stabilization parts of the discriminator diagram, the blue block is labeled “New toRGB.” However, since this is the discriminator, the layer appears to be converting an RGB image into feature maps. Therefore, it seems that the label should be “New fromRGB” rather than “New toRGB.”
My understanding is that:
Because Figure 10-4 describes the discriminator, I believe the blue block should be labeled “New fromRGB.”
Thank you for your excellent book. I am learning a lot from it, and I wanted to share this possible correction respectfully.
Sincerely,
Basil Aljarih