3 ms·
Famously, some of the image models were so overtrained they could still yield impressive results if the colors were removed.
by fbdab103 3y ago
Famously, some of the image models were so overtrained they could still yield impressive results if the colors were removed.
- lumost 3y agoThis wasn't so much overtraining, as the models learning something different than what we expected. If you look at a pixel by pixel representation of an image, textures tend to be more significant/unique patterns than shapes. There are some funny studies from the mid 2010s exploring this.