Research map / Image generation / GANs
Stabilized GAN training
Improve GAN stability, convergence, and limited-data training with new objectives and discriminators.
144 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
Still moving The best reported result is still improving on 1 of 6 tracked benchmarks.
24 papers in 2023–24 and 9 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic shrank, from 41.4% to 36.0%.
Open questions the data points to
- CIFAR-10 unconditional (FID): best went from 1.54 (up to 2024) to 1.38 in 2025 — Direct Discriminative Optimization: Your Likelihood-Based Visual Generative Model is Secretly a GAN Discriminator (ICML 2025 · 0 citations)
- Papers report WGAN-GP on CIFAR-10 (FID) anywhere from 18.86 to 40.2: comparisons against it depend on whose number you take.
- Papers report WGAN-GP on CIFAR-10 (Inception Score) anywhere from 6.68 to 7.86: comparisons against it depend on whose number you take.
No longer reported by papers here: CelebA (FID, last in 2023); STL-10 (FID, last in 2024); CIFAR-10 (Inception score, last in 2021); FFHQ (FID, last in 2023).
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 · SeqGAN: Sequence Generative Adversarial Nets with Policy Gradient (AAAI 2017 · 2,289 citations)
- 2018 · Spectral Normalization for Generative Adversarial Networks (ICLR 2018 · 3,438 citations)
- 2019 · Self-Attention Generative Adversarial Networks (ICML 2019 · 2,555 citations)
- 2020 · Training Generative Adversarial Networks with Limited Data (NeurIPS 2020 · 931 citations)
- 2021 · Improved Consistency Regularization for GANs (AAAI 2021 · 105 citations)
- 2022 · GenCo: Generative Co-training for Generative Adversarial Networks with Limited Data (AAAI 2022 · 30 citations)
- 2023 · Diffusion-GAN: Training GANs with Diffusion (ICLR 2023 · 66 citations)
- 2024 · Generative Model Perception Rectification Algorithm for Trade-Off between Diversity and Quality (AAAI 2024 · 12 citations)
- 2025 · MCGAN: Enhancing GAN Training with Regression-Based Generator Loss (AAAI 2025 · 2 citations)
- 2026 · Verification of the Implicit World Model in a Generative Model via Adversarial Sequences (ICLR 2026 · 0 citations)
Earliest papers here
- SeqGAN: Sequence Generative Adversarial Nets with Policy Gradient (AAAI 2017 · 2,289 citations)
- Improved Training of Wasserstein GANs (NeurIPS 2017 · 1,476 citations)
- VEEGAN: Reducing Mode Collapse in GANs using Implicit Variational Learning (NeurIPS 2017 · 495 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)
24 papers, 2018–2025 · one report since 2024, not better
| Year | Best so far | Set by |
|---|---|---|
| 2018 | 16.43 | Banach Wasserstein GAN (NeurIPS 2018 · 121 citations) |
| 2019 | 14.7 | Progressive Augmentation of GANs (NeurIPS 2019 · 16 citations) |
| 2021 | 12.1 | Dynamically Grown Generative Adversarial Networks (AAAI 2021 · 13 citations) |
| 2023 | 10.54 | Dynamically Masked Discriminator for GANs (NeurIPS 2023 · 0 citations) |
| 2024 | 2.06 | SMaRt: Improving GANs with Score Matching Regularity (ICML 2024 · 0 citations) |
CelebA · FID (lower is better)
10 papers, 2018–2023 · not reported since 2023
| Year | Best so far | Set by |
|---|---|---|
| 2018 | 9.3 | Coulomb GANs: Provably Optimal Nash Equilibria via Potential Fields (ICLR 2018 · 45 citations) |
| 2019 | 6.79 | Improving MMD-GAN Training with Repulsive Loss Function (ICLR 2019 · 42 citations) |
| 2022 | 6.3 | DO-GAN: A Double Oracle Framework for Generative Adversarial Networks (CVPR 2022 · 2 citations) |
CIFAR-10 unconditional · FID (lower is better)
8 papers, 2019–2025 · still improving
| Year | Best so far | Set by |
|---|---|---|
| 2019 | 18.58 | On Self Modulation for Generative Adversarial Networks (ICLR 2019 · 73 citations) |
| 2020 | 12.2 | CircleGAN: Generative Adversarial Learning across Spherical Circles (NeurIPS 2020 · 1 citation) |
| 2021 | 11.82 | Adaptive Weighted Discriminator for Training Generative Adversarial Networks (CVPR 2021 · 14 citations) |
| 2023 | 2.55 | CoopInit: Initializing Generative Adversarial Networks via Cooperative Learning (AAAI 2023 · 2 citations) |
| 2024 | 1.54 | Structure-Guided Adversarial Training of Diffusion Models (CVPR 2024 · 10 citations) |
| 2025 | 1.38 | Direct Discriminative Optimization: Your Likelihood-Based Visual Generative Model is Secretly a GAN Discriminator (ICML 2025 · 0 citations) |
STL-10 · FID (lower is better)
7 papers, 2018–2024 · not reported since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2018 | 40.1 | Spectral Normalization for Generative Adversarial Networks (ICLR 2018 · 3,438 citations) |
| 2019 | 30.8 | Improving GAN with Neighbors Embedding and Gradient Matching (AAAI 2019 · 14 citations) |
| 2020 | 26.16 | A Characteristic Function Approach to Deep Implicit Generative Modeling (CVPR 2020 · 30 citations) |
| 2022 | 23 | Improving Generative Adversarial Networks via Adversarial Learning in Latent Space (NeurIPS 2022 · 1 citation) |
| 2024 | 19.27 | Discriminative Forests Improve Generative Diversity for Generative Adversarial Networks (AAAI 2024 · 3 citations) |
CIFAR-10 · Inception score (higher is better)
6 papers, 2018–2021 · not reported since 2021
| Year | Best so far | Set by |
|---|---|---|
| 2018 | 8.8 | Progressive Growing of GANs for Improved Quality, Stability, and Variation (ICLR 2018 · 1,535 citations) |
FFHQ · FID (lower is better)
6 papers, 2020–2023 · not reported since 2023
| Year | Best so far | Set by |
|---|---|---|
| 2020 | 7.48 | A U-Net Based Discriminator for Generative Adversarial Networks (CVPR 2020 · 350 citations) |
| 2021 | 4.06 | SSD-GAN: Measuring the Realness in the Spatial and Spectral Domains (AAAI 2021 · 42 citations) |
| 2023 | 2.45 | Spider GAN: Leveraging Friendly Neighbors To Accelerate GAN Training (CVPR 2023 · 2 citations) |
What has been tried
134 papers here report 1,464 measured changes (a component added, removed or swapped, compared with a baseline): helped 71%, no clear effect 6%, hurt 19%, mixed 4%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| Remove informative weights from DMPRA | CelebA-HQ · FID | Hurt | Generative Model Perception Rectification Algorithm for Trade-Off between Diversity and Quality (AAAI 2024 · 12 citations) | |
| Add structural guidance | ImageNet · FID | 4.59 -> 3.57 | Helped | Structure-Guided Adversarial Training of Diffusion Models (CVPR 2024 · 10 citations) |
| Use 8 × 8 for unconditional training resolution | AnimalFaces-LT · FID-FS | 50.3 -> 48.4 | Helped | Taming the Tail in Class-Conditional GANs: Knowledge Sharing via Unconditional Training at Lower Resolutions (CVPR 2024 · 8 citations) |
| Increase LoRA updates to 50 steps per iteration | 2D score distillation · image quality | Hurt | Adversarial Score Distillation: When score distillation meets GAN (CVPR 2024 · 7 citations) | |
| Increase K from 1 to 10 while keeping the total discriminator parameters at 2M | STL10 · FID | 29.10 -> 22.72 | Helped | Discriminative Forests Improve Generative Diversity for Generative Adversarial Networks (AAAI 2024 · 3 citations) |
| Vary q1 across the tested values | Steering Angle 64×64 · SFID | 1.125 (0.510); 1.114 (0.503); 1.161 (0.540); 1.187 (0.558) | No clear effect | Turning Waste into Wealth: Leveraging Low-Quality Samples for Enhancing Continuous Conditional Generative Adversarial Networks (AAAI 2024 · 3 citations) |
| add regression generator loss without DiffAug | CIFAR-10 · FID | 5.31 -> 4.51 | Hurt | MCGAN: Enhancing GAN Training with Regression-Based Generator Loss (AAAI 2025 · 2 citations) |
| remove activation functions, making score functions and discriminators purely linear | cheetah-run · performance | Mixed | Diffusing States and Matching Scores: A New Framework for Imitation Learning (ICLR 2025 · 1 citation) | |
| Add 0-mean regularization | CIFAR-10 10% data · tFID | 9.20 -> 8.27 | Helped | CHAIN: Enhancing Generalization in Data-Efficient GANs via lipsCHitz continuity constrAIned Normalization (CVPR 2024 · 1 citation) |
| Set λ to 0.25 rather than λ=0.00 | Obama 100-shot · FID | 46.87 -> 41.27 | Helped | Improving the Training of Data-Efficient GANs via Quality Aware Dynamic Discriminator Rejection Sampling (CVPR 2025 · 0 citations) |
| Use a wide range of α, β settings | Class-conditional CIFAR-10 · FID | Helped | Direct Discriminative Optimization: Your Likelihood-Based Visual Generative Model is Secretly a GAN Discriminator (ICML 2025 · 0 citations) | |
| Apply MS^3D to both real and generated data instead of real data only | OxfordDog · FID | 47.05 -> 42.73 | Helped | MS$^3$D: A RG Flow-Based Regularization for GAN Training with Limited Data (ICML 2024 · 0 citations) |
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 | 18.86 – 40.2 | 6 papers, e.g. ICML 2019 and ICLR 2018 |
| WGAN-GP | CIFAR-10 | Inception Score | 6.68 – 7.86 | 4 papers, e.g. ICLR 2018 and AAAI 2021 |
| WGAN-GP | CelebA | FID | 10.03 – 54.71 | 2 papers, e.g. CVPR 2020 and NeurIPS 2019 |
| WGAN-GP | STL10 | FID | 27.85 – 55.1 | 2 papers, e.g. CVPR 2020 and ICLR 2018 |
| DistGAN | CIFAR-10 | FID | 17.61 – 22.95 | 2 papers, e.g. AAAI 2021 and AAAI 2019 |
What reviewers will expect
Datasets and baselines named in the 6 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- CIFAR-10 · 2 of 6 papers
Baselines
- StyleGAN2-ada · 2 of 6 papers
- EDM · 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 33 papers; from OpenAlex and the papers' affiliations).
- Tencent · 4 papers
- Tsinghua University · 3 papers
- Stevens Institute of Technology · 2 papers
- King Abdullah University of Science and Technology · 2 papers
- Ant Group · 2 papers
- Google · 2 papers
- Independent Researcher · 2 papers
- Australian National University · 2 papers
- University of Texas at Austin · 2 papers
Use it
Download BibTeX of 40 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
- GAN latent optimization and evaluation (52 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).