Artificial Intelligence For the Stereographer: Links
Watch the workshop by Gordon Au here; originally presented at NSA Virtual 3D-Con 2021
Many modern developments in AI deal with the 3rd dimension, from self-driving cars that perceive and navigate their environments in 3D, to programs that can build 3D models of scenes simply by analyzing 2D videos of them. Though some of these AIs are only accessible to researchers and tech companies, some AI applications like upscaling images or colorizing black and white photos are immediately useful and available to all. This workshop will show you how to use AIs of special interest to stereographers, via the free tool Google Colab, to do things like creating depth maps, stereopairs, animations, and stereovideos—even from 2D images.
If you encounter errors using the Colab Notebooks below, please leave a comment on the YouTube video above, or contact me ">here.
- Recommended: BigJPG.com
- “Neural Image Super-Resolution” by Christian Ledig et al. *not covered in this workshop
- On Google Colab
Colorizing Black & White Images
- Website version: DeepAI
- DeOldify by Jason Antic et al.
- Google Colab ‘artistic version’ (for more interesting details, and vibrance): direct upload version / original
Google Colab ‘stable version’ (better for landscapes and portraits)
Creating Depth Maps From 2D Images
- “Boosting Monocular Depth Estimation Models to High-Resolution via Content-Adaptive Multi-Resolution Merging” by S. Mahdi H. Miangoleh et al.
- On Google Colab
Bug fix version (if the above gives you an error)
- MiDaS: “Towards Robust Monocular Depth Estimation: Mixing Datasets for Zero-shot Cross-dataset Transfer” by René Ranftl et al.
- On Google Colab: direct upload version / original
Using Depth Maps to Animate 2D Images
- 3D-Photo-Inpainting: “3D Photography using Context-aware Layered Depth Inpainting” by Meng-Li Shih et al.
- On Google Colab Fixed!—[*NOTE* currently not working due to missing files. I’ve reported the issue to the authors.]
Extrapolating Depth Maps From 2D Videos
- “Consistent Video Depth Estimation” by Xuan Lo et al.
On Google Colab *NEW / not covered in this workshop
- “Neural Scene Flow Fields for Space-Time View Synthesis of Dynamic Scenes” by Zhengqi Li et al.