Research map / Image generation / GANs
GAN latent optimization and evaluation
Improve generators via latent optimization, pretrained GANs, optimal transport, and divergence-based evaluation.
52 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2017–2026) · an approach in GANs, keyword cluster “gans · mode · samples”
State History Benchmarks Tried Baselines Checklist Preprints Who Export
Where it stands
Few recent reports None of the tracked benchmarks has a report from 2025–26.
6 papers in 2023–24 and 3 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic grew, from 10.3% to 12.0%.
Open questions the data points to
- Papers report WGAN-GP on CIFAR-10 (FID) anywhere from 24.8 to 72.23: comparisons against it depend on whose number you take.
- Papers report WGAN-GP on CIFAR-10 (Inception Score) anywhere from 4.93 to 7.36: comparisons against it depend on whose number you take.
No longer reported by papers here: CIFAR-10 (FID, last in 2024); CIFAR-10 (IS, last in 2024); CelebA (FID, last in 2021); CIFAR10 (Inception score, last in 2019).
How this is decided
A benchmark is tracked when at least four papers of this approach report their own method on the same dataset and metric in at least two different years. Its record is the best number reported up to 2024 compared with the best reported in 2025–26: still improving means at least 1% better, flat means at most 0.2% better with at least two papers from 2025–26 trying. Numbers on a different scale from the rest (a fraction among percentages) and far outliers are left out. Results come from different papers with different setups, so they are signals, not head-to-head comparisons. The share of papers is counted at ICML, NeurIPS, ICLR, CVPR and AAAI.
How it got here
Most cited paper of each year
- 2017 · Triple Generative Adversarial Nets (NeurIPS 2017 · 224 citations)
- 2018 · Optimizing the Latent Space of Generative Networks (ICML 2018 · 196 citations)
- 2019 · GAN Dissection: Visualizing and Understanding Generative Adversarial Networks (ICLR 2019 · 179 citations)
- 2020 · MSG-GAN: Multi-Scale Gradients for Generative Adversarial Networks (CVPR 2020 · 245 citations)
- 2021 · GAN Ensemble for Anomaly Detection (AAAI 2021 · 72 citations)
- 2022 · Ensembling Off-the-Shelf Models for GAN Training (CVPR 2022 · 85 citations)
- 2023 · Diffusion Models Beat GANs on Topology Optimization (AAAI 2023 · 72 citations)
- 2024 · RG-GAN: Dynamic Regenerative Pruning for Data-Efficient Generative Adversarial Networks (AAAI 2024 · 14 citations)
- 2025 · Diverse Rare Sample Generation with Pretrained GANs (AAAI 2025 · 0 citations)
- 2026 · When and Why Adversarial Training Improves PINNs: A Neural Tangent Kernel Perspective (NeurIPS 2026 · 0 citations)
Earliest papers here
- Triple Generative Adversarial Nets (NeurIPS 2017 · 224 citations)
- Stabilizing Training of Generative Adversarial Networks through Regularization (NeurIPS 2017 · 212 citations)
- Generalization and Equilibrium in Generative Adversarial Nets (GANs) (ICML 2017 · 199 citations)
Benchmark records
The best number each year that beat every earlier one, among papers of this approach reporting their own method. Numbers are extracted by AI from each paper's text; check the paper before citing.
CIFAR-10 · FID (lower is better)
9 papers, 2018–2024 · not reported since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2018 | 27.4 | First Order Generative Adversarial Networks (ICML 2018 · 6 citations) |
| 2019 | 12.47 | Quality Aware Generative Adversarial Networks (NeurIPS 2019 · 20 citations) |
| 2024 | 3.16 | Consistency-GAN: Training GANs with Consistency Model (AAAI 2024 · 4 citations) |
CIFAR-10 · IS (higher is better)
6 papers, 2018–2024 · not reported since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2018 | 6.761 | Quantitatively Evaluating GANs With Divergences Proposed for Training (ICLR 2018 · 13 citations) |
| 2019 | 8.37 | Quality Aware Generative Adversarial Networks (NeurIPS 2019 · 20 citations) |
| 2020 | 8.63 | MSG-GAN: Multi-Scale Gradients for Generative Adversarial Networks (CVPR 2020 · 245 citations) |
| 2024 | 9.76 | Consistency-GAN: Training GANs with Consistency Model (AAAI 2024 · 4 citations) |
CelebA · FID (lower is better)
6 papers, 2018–2021 · not reported since 2021
| Year | Best so far | Set by |
|---|---|---|
| 2018 | 6 | First Order Generative Adversarial Networks (ICML 2018 · 6 citations) |
| 2021 | 5.98 | Self-Diagnosing GAN: Diagnosing Underrepresented Samples in Generative Adversarial Networks (NeurIPS 2021 · 14 citations) |
CIFAR10 · Inception score (higher is better)
5 papers, 2017–2019 · not reported since 2019
| Year | Best so far | Set by |
|---|---|---|
| 2017 | 7.72 | Generalization and Equilibrium in Generative Adversarial Nets (GANs) (ICML 2017 · 199 citations) |
What has been tried
50 papers here report 559 measured changes (a component added, removed or swapped, compared with a baseline): helped 69%, no clear effect 7%, hurt 19%, mixed 5%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| add recurrent regeneration to dynamic pruning | CIFAR-10 · FID | Helped | RG-GAN: Dynamic Regenerative Pruning for Data-Efficient Generative Adversarial Networks (AAAI 2024 · 14 citations) | |
| Add discriminator pruning | Cityscapes · mIoU | 39.37 -> 39.84 | Helped | Compressing Image-to-Image Translation GANs Using Local Density Structures on Their Learned Manifold (AAAI 2024 · 11 citations) |
| Inject instance noise through the consistency mapping module during adversarial training | CIFAR-10 · training stability | Helped | Consistency-GAN: Training GANs with Consistency Model (AAAI 2024 · 4 citations) | |
| Add similarity regularization to the rarity-only objective | FFHQ · RS | 18.99 -> 21.11 | Helped | Diverse Rare Sample Generation with Pretrained GANs (AAAI 2025 · 0 citations) |
| Add similarity-guided training | FFHQ → Sunglasses (10-shot) · FID | 41.88 -> 26.41 | Helped | Bridging Data Gaps in Diffusion Models with Adversarial Noise-Based Transfer Learning (ICML 2024) |
Baseline numbers that disagree
The same baseline on the same benchmark and metric, reported more than 5% apart by different papers here. Training budgets, data splits and reimplementations differ, so state which number you compare with.
| Baseline | Benchmark | Metric | Reported | Papers |
|---|---|---|---|---|
| WGAN-GP | CIFAR-10 | FID | 24.8 – 72.23 | 5 papers, e.g. ICML 2018 and NeurIPS 2019 |
| WGAN-GP | CIFAR-10 | Inception Score | 4.93 – 7.36 | 3 papers, e.g. NeurIPS 2019 and AAAI 2019 |
| BEGAN | CIFAR10 | FID | 32.3 – 71.4 | 2 papers, e.g. AAAI 2019 and CVPR 2019 |
| BEGAN | CelebA | FID | 28.5 – 38.9 | 2 papers, e.g. ICML 2018 and CVPR 2019 |
| WGAN | CIFAR10 | FID | 42.8 – 55.2 | 2 papers, e.g. AAAI 2019 and CVPR 2019 |
What reviewers will expect
Datasets and baselines named in the 6 papers of this approach from 2024–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- FFHQ · 3 of 6 papers
- CIFAR-10 · 2 of 6 papers
- ImageNet · 2 of 6 papers
Baselines
- StyleGAN2 · 2 of 6 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 9 papers; from OpenAlex and the papers' affiliations).
- The Hong Kong Polytechnic University · 2 papers
Use it
Download BibTeX of 27 key papersOpen the topic in the interactive map
Get alerts. Describe your idea in scime atlas and save it (free scime account): you get a weekly email when new preprints land close to it.
Related approaches
Same keyword cluster
- Stabilized GAN training (144 papers)
- GANs for text generation (21 papers)
- Generative adversarial network theory (62 papers)
Elsewhere in the topic
- Large-scale GAN latent analysis (108 papers)
- GAN-based image manipulation (90 papers)
- Conditional GANs (25 papers)
- GAN vocoders and audio synthesis (13 papers)
Benchmark numbers, ablations and baseline numbers are extracted by AI from each paper's full text and can be wrong; the paper is the authority. Data: results index of 2026-11-02, citation counts from OpenAlex (2026-10-09).