3 ms·
A while back I did a saliency map by moving a square pixel all across a face image & comparing the embedding to the orginal image. Turns out it is just the T sh
by villgax 3y ago
A while back I did a saliency map by moving a square pixel all across a face image & comparing the embedding to the orginal image. Turns out it is just the T shape of our face that had the largest impact. Not the cheeks or forehead.
- Daub 3y agoThis comes as no surprise and is pretty much what I teach my drawing students. nice to see it confirmed. Did you publish this? When drawing, novice artists give this T shape the maximum attention to the exclsioon of all else.
- villgax 3y agoDidn't publish in public, I'll see if I can recreate it.