What it does This repository is the living supplement to a 2022 TPAMI survey on GAN inversion — the art of taking a real image and finding its coordinates inside a trained generator’s latent space. The authors maintain a categorized bibliography of pretrained models (2D and 3D-aware), inversion methods, latent-space editing techniques, and downstream applications from face recognition to medical imaging.

The interesting bit The list has quietly expanded beyond its original scope: it now tracks diffusion inversion and diffusion latent editing too, acknowledging that the same “find the hidden representation” problem outlived the GAN era. There’s even a helper script to auto-generate markdown entries from arXiv, which suggests the maintainers expect this literature to keep accelerating.

Key highlights

• Covers the full stack: ProGAN → StyleGAN3, EG3D, pi-GAN, plus encoders like TriPlaneNet and GOAE
• Sections on 3D inversion methods are distinct from 2D, with separate tracks for general techniques and application-specific hacks
• Includes diffusion inversion as a parallel thread, not an afterthought
• Applications span image restoration, 3D reconstruction, compressed sensing, fairness, and security
• PRs welcome; the authors provide an arXiv-to-markdown script for contributors

Caveats

• The README is a reference list, not runnable code — you’ll go elsewhere for implementations
• Some newer entries appear to be paper-only (no code links yet)

Verdict Bookmark this if you’re building anything that needs to project real images into generative latent spaces, or if you’re writing a literature review and want to avoid missing a baseline. Skip it if you’re looking for a drop-in inversion library — this is the map, not the territory.