Animated 3D photos generated from single image with AI:
[CVPR 2020] 3D Photography using Context-aware Layered Depth Inpainting https://github.com/vt-vl-lab/3d-photo-inpainting
2-11 animated GIF 1,12-22 static stereograph at higher res 18: Handcrafted by not me 19: AI generated 20: Handcrafted by not me 21: AI generated
How to view stereogragh: 1. While focusing on the image, place ONE upward pointing finger slightly above your nose, your brain should then see TWO "phantom fingers" 2. While still focusing on the image, adjust position of your real finger, till the tip of the two phantom fingers point to the character face at the left half stereograph, and the character face at the right half stereograph, respectively. 3. Now focus on your finger, the phantom fingers coincide and you see ONE finger only now. The character face from left/right half of the sterograph also coincide, forming the 3D effect image behind your finger 4. Move your focus from your finger back to the 3D effect image behind your finger. Done
Try on the static one first, before moving on to the animated GIF
Fun Fact: The same trick also lets you steamroll those spot the difference between left/right image games.
Comment of successfulness: hA: Good! Best so far, except that the left face and hair decoration is off hB: Fair. Depth estimation fails at dick hC: Fair. Overall a bit flat. Problem with hair hD: Fail. Discontinuity of depth at left hand and right leg hE: Good! hF: Fair. Problem with hair again. Weird discontinuity at dick/right leg boundary hG: Fail. Just Fail hH: Good! Depth estimate good. Fails at filling of background. hI: Fair. Discontinuity of depth at shoulder due to hair obstruction. hJ: Fair. A bit weird at the belly.
Summary on the AI: Works well on round things ( ಠ ͜ʖಠ) Works well when the original image do good lighting and shadings. Struggles with hair. Sometimes the depth difference between body parts is too large.
Comments on result quality / tech welcomed. Hope others would pick up and make more good stuffs.