Research map / Image generation / Editing and style transfer
Pose-guided person image synthesis
Generate person images in new poses using flows, texture, and diffusion.
29 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2017–2026) · an approach in Editing and style transfer, keyword cluster “translation · style transfer · image image”
State History Benchmarks Tried Baselines Checklist Preprints Export
Where it stands
Still moving The best reported result is still improving on 2 of 6 tracked benchmarks.
5 papers in 2023–24 and 5 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic shrank, from 1.8% to 1.3%.
Open questions the data points to
- DeepFashion 256×176 (LPIPS): best went from 0.128 (up to 2024) to 0.114 in 2025 — Learning Flow Fields in Attention for Controllable Person Image Generation (CVPR 2025 · 12 citations)
- DeepFashion 256×176 (FID): best went from 5.874 (up to 2024) to 5.71 in 2025 — Learning Flow Fields in Attention for Controllable Person Image Generation (CVPR 2025 · 12 citations)
- Papers report DIST on DeepFashion (FID) anywhere from 7.629 to 10.84: comparisons against it depend on whose number you take.
No longer reported by papers here: DeepFashion (SSIM, last in 2024); DeepFashion (FID, last in 2024); DeepFashion (IS, last in 2022).
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 · Pose Guided Person Image Generation (NeurIPS 2017 · 676 citations)
- 2018 · Disentangled Person Image Generation (CVPR 2018 · 575 citations)
- 2019 · Progressive Pose Attention Transfer for Person Image Generation (CVPR 2019 · 385 citations)
- 2020 · Controllable Person Image Synthesis With Attribute-Decomposed GAN (CVPR 2020 · 269 citations)
- 2021 · PISE: Person Image Synthesis and Editing With Decoupled GAN (CVPR 2021 · 151 citations)
- 2022 · Feature Erasing and Diffusion Network for Occluded Person Re-Identification (CVPR 2022 · 203 citations)
- 2023 · Person Image Synthesis via Denoising Diffusion Model (CVPR 2023 · 127 citations)
- 2024 · IMAGPose: A Unified Conditional Framework for Pose-Guided Person Generation (NeurIPS 2024 · 51 citations)
- 2025 · Learning Flow Fields in Attention for Controllable Person Image Generation (CVPR 2025 · 12 citations)
- 2026 · PHAC: Promptable Human Amodal Completion (CVPR 2026 · 0 citations)
Earliest papers here
- Pose Guided Person Image Generation (NeurIPS 2017 · 676 citations)
- Disentangled Person Image Generation (CVPR 2018 · 575 citations)
- FD-GAN: Pose-guided Feature Distilling GAN for Robust Person Re-identification (NeurIPS 2018 · 298 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.
DeepFashion · SSIM (higher is better)
9 papers, 2017–2024 · not reported since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2017 | 0.762 | Pose Guided Person Image Generation (NeurIPS 2017 · 676 citations) |
| 2018 | 0.793 | Soft-Gated Warping-GAN for Pose-Guided Person Image Synthesis (NeurIPS 2018 · 39 citations) |
DeepFashion · FID (lower is better)
6 papers, 2019–2024 · not reported since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2019 | 29.68 | PCGAN: Partition-Controlled Human Image Generation (AAAI 2019 · 20 citations) |
| 2020 | 10.57 | Deep Image Spatial Transformation for Person Image Generation (CVPR 2020 · 216 citations) |
| 2021 | 8.755 | Structure-aware Person Image Generation with Pose Decomposition and Semantic Correlation (AAAI 2021 · 21 citations) |
| 2024 | 5.874 | IMAGPose: A Unified Conditional Framework for Pose-Guided Person Generation (NeurIPS 2024 · 51 citations) |
DeepFashion · IS (higher is better)
6 papers, 2017–2022 · not reported since 2022
| Year | Best so far | Set by |
|---|---|---|
| 2017 | 3.09 | Pose Guided Person Image Generation (NeurIPS 2017 · 676 citations) |
| 2018 | 3.314 | Soft-Gated Warping-GAN for Pose-Guided Person Image Synthesis (NeurIPS 2018 · 39 citations) |
| 2019 | 3.536 | PCGAN: Partition-Controlled Human Image Generation (AAAI 2019 · 20 citations) |
| 2022 | 3.632 | Self-Supervised Correlation Mining Network for Person Image Generation (CVPR 2022 · 15 citations) |
DeepFashion 256×176 · SSIM (higher is better)
6 papers, 2022–2025 · one report since 2024, not better
| Year | Best so far | Set by |
|---|---|---|
| 2022 | 0.718 | Neural Texture Extraction and Distribution for Controllable Person Image Synthesis (CVPR 2022 · 86 citations) |
| 2023 | 0.731 | Person Image Synthesis via Denoising Diffusion Model (CVPR 2023 · 127 citations) |
| 2024 | 0.756 | IMAGPose: A Unified Conditional Framework for Pose-Guided Person Generation (NeurIPS 2024 · 51 citations) |
DeepFashion 256×176 · LPIPS (lower is better)
6 papers, 2022–2025 · still improving
| Year | Best so far | Set by |
|---|---|---|
| 2022 | 0.175 | Neural Texture Extraction and Distribution for Controllable Person Image Synthesis (CVPR 2022 · 86 citations) |
| 2023 | 0.168 | Person Image Synthesis via Denoising Diffusion Model (CVPR 2023 · 127 citations) |
| 2024 | 0.128 | IMAGPose: A Unified Conditional Framework for Pose-Guided Person Generation (NeurIPS 2024 · 51 citations) |
| 2025 | 0.114 | Learning Flow Fields in Attention for Controllable Person Image Generation (CVPR 2025 · 12 citations) |
DeepFashion 256×176 · FID (lower is better)
6 papers, 2022–2025 · still improving
| Year | Best so far | Set by |
|---|---|---|
| 2022 | 8.684 | Neural Texture Extraction and Distribution for Controllable Person Image Synthesis (CVPR 2022 · 86 citations) |
| 2023 | 6.367 | Person Image Synthesis via Denoising Diffusion Model (CVPR 2023 · 127 citations) |
| 2024 | 5.874 | IMAGPose: A Unified Conditional Framework for Pose-Guided Person Generation (NeurIPS 2024 · 51 citations) |
| 2025 | 5.71 | Learning Flow Fields in Attention for Controllable Person Image Generation (CVPR 2025 · 12 citations) |
What has been tried
28 papers here report 295 measured changes (a component added, removed or swapped, compared with a baseline): helped 76%, no clear effect 3%, hurt 19%, mixed 1%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| add CVA to B0 | DeepFashion · SSIM | 0.7235 -> 0.7281 | Helped | IMAGPose: A Unified Conditional Framework for Pose-Guided Person Generation (NeurIPS 2024 · 51 citations) |
| Use PRD prompt with query-bias HGA rather than B1's multi-scale appearance prompt and K,V bias | DeepFashion 256×176 · LPIPS | 0.2018 -> 0.1519 | Helped | Coarse-to-Fine Latent Diffusion for Pose-Guided Person Image Synthesis (CVPR 2024 · 40 citations) |
| add the refining conditional diffusion model | DeepFashion 256×176 · SSIM | 0.7357 -> 0.7444 | Helped | Advancing Pose-Guided Image Synthesis with Progressive Conditional Diffusion Models (ICLR 2024 · 17 citations) |
| Add Leffa attention-flow regularization | VITON-HD paired and unpaired · FID | 5.31 -> 4.54; 9.38 -> 8.52 | Helped | Learning Flow Fields in Attention for Controllable Person Image Generation (CVPR 2025 · 12 citations) |
| add graph matching-guided neural representation learning | GOD · CLIPT@50 | 12.0% -> 24.0% | Helped | Mind Artist: Creating Artistic Snapshots with Human Thought (CVPR 2024 · 7 citations) |
| Train on synthetic pairs instead of segmented pairs | Multi-garment human image generation · MP-LPIPS | 0.374 -> 0.197 | Helped | Controllable Human Image Generation with Personalized Multi-Garments (CVPR 2025 · 4 citations) |
| Add Global-aware Pose Generation | UBC Fashion · SSIM | 0.9362 -> 0.9411 | Helped | Disentangled Pose and Appearance Guidance for Multi-Pose Generation (CVPR 2025 · 2 citations) |
| Use two reference images instead of the single-reference baseline | MSTed · FVD | 26.32 vs. 9.82 | Helped | Free-viewpoint Human Animation with Pose-correlated Reference Selection (CVPR 2025 · 1 citation) |
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 |
|---|---|---|---|---|
| DIST | DeepFashion | FID | 7.629 – 10.84 | 2 papers, e.g. AAAI 2021 and CVPR 2022 |
| DIST | Market-1501 | FID | 19.74 – 21.54 | 2 papers, e.g. CVPR 2022 and AAAI 2021 |
What reviewers will expect
Datasets and baselines named in the 8 papers of this approach from 2024–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- DeepFashion · 4 of 8 papers
- VITON-HD · 2 of 8 papers
- DressCode · 2 of 8 papers
- Market-1501 · 2 of 8 papers
Baselines
- PIDM · 5 of 8 papers
- NTED · 5 of 8 papers
- DPTN · 4 of 8 papers
- PoCoLD · 3 of 8 papers
- CASD · 3 of 8 papers
- GFLA · 3 of 8 papers
- PISE · 3 of 8 papers
- CFLD · 3 of 8 papers
Recent preprints
Loading recent arXiv preprints…
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
- Image-to-image translation (127 papers)
- Texture and sketch synthesis (130 papers)
- Arbitrary neural style transfer (124 papers)
- Virtual try-on (46 papers)
Elsewhere in the topic
- Instruction-based diffusion editing (151 papers)
- Inversion-based diffusion editing (113 papers)
- Identity-preserving face manipulation (88 papers)
- Localized and attribute-specific editing (73 papers)
- Diffusion face generation and editing (61 papers)
- GAN inversion for editing (25 papers)
- Talking face generation (21 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).