Research map / Image generation / GANs
SinGAN: Learning a Generative Model From a Single Natural Image
Tamar Rott Shaham, Tali Dekel, Tomer Michaeli
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| start generation from scale N-1 instead of the coarsest scale N (paired) (ablation) | SinGAN · AMT Real/Fake paired | confusion rate: 21.45% -> 30.45% | Improved |
| start generation from scale N-1 instead of N (unpaired) (ablation) | SinGAN · AMT Real/Fake unpaired | confusion rate: 42.9% -> 47.04% | Improved |
| start generation from scale N-1 instead of N (ablation) | SinGAN · single-image SIFID | SIFID: 0.09 -> 0.05 | Improved |
| start generation from scale N-1 instead of N (ablation) | SinGAN · sample diversity | normalized pixel std: 0.5 -> 0.35 | Hurt |
| train with more pyramid scales (ablation) | SinGAN · qualitative single images | global structure preservation | Improved |
| use SinGAN (internal, single image) instead of ZSSR for super-resolution | internal SR methods · BSD100 super-resolution | NIQE: 7.13 -> 3.71 | Improved |
| use SinGAN instead of DIP for super-resolution | internal SR methods · BSD100 super-resolution | NIQE: 6.35 -> 3.71 | Improved |
| use SinGAN instead of ZSSR for super-resolution | internal SR methods · BSD100 super-resolution | RMSE: 13.08 -> 16.22 | Hurt |
| use SinGAN instead of EDSR for super-resolution | SR methods · BSD100 super-resolution | NIQE: 6.50 -> 3.71 | Improved |
| use SinGAN instead of SRGAN for super-resolution | SR methods · BSD100 super-resolution | NIQE: 3.41 vs 3.71 | No clear effect |
Later papers that compared with this method
13 later accepted papers report SinGAN as a baseline in their results tables; 13 of them report a better number on most of the results they share with it.
- SinDiffusion: Learning a Diffusion Model From a Single Natural Image (TPAMI 2025 · better on 2 of 2 shared results)
- SinDDM: A Single Image Denoising Diffusion Model (ICML 2023 · better on 5 of 6 shared results)
- Neural Texture Synthesis With Guided Correspondence (CVPR 2023 · better on 1 of 1 shared result)
- Generating Natural Images with Direct Patch Distributions Matching (ECCV 2022 · better on 4 of 4 shared results)
- InfinityGAN: Towards Infinite-Pixel Image Synthesis (ICLR 2022 · better on 2 of 3 shared results)
- GLEAN: Generative Latent Bank for Image Super-Resolution and Beyond (TPAMI 2022 · better on 1 of 1 shared result)
- Patchwise Generative ConvNet: Training Energy-Based Models From a Single Natural Image for Internal Learning (CVPR 2021 · better on 5 of 8 shared results)
- SinIR: Efficient General Image Manipulation with Single Image Reconstruction (ICML 2021 · better on 3 of 4 shared results)
5 more not listed.
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- PetsGAN: Rethinking Priors for Single Image Generation (AAAI 2022 · similarity 0.92)
- Drop the GAN: In Defense of Patches Nearest Neighbors As Single Image Generative Models (CVPR 2022 · similarity 0.90)
- SinGRAF: Learning a 3D Generative Radiance Field for a Single Scene (CVPR 2023 · similarity 0.89)
- InGAN: Capturing and Retargeting the "DNA" of a Natural Image (ICCV 2019 · similarity 0.89)
- DeLiGAN : Generative Adversarial Networks for Diverse and Limited Data (CVPR 2017 · similarity 0.89)
- Large Scale GAN Training for High Fidelity Natural Image Synthesis (ICLR 2019 · similarity 0.89)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
