{"url":"/sota/image-generation-on-celeba-64x64","task":{"name":"Image Generation","url":"/task/image-generation","note":null},"dataset":{"name":"CelebA 64x64","url":"/dataset/celeba"},"category":"Computer Vision","categories":["Computer Vision","Medical","Miscellaneous","Natural Language Processing"],"category_note":null,"description":"**Image Generation** (synthesis) is the task of generating new images from an existing dataset.\r\n\r\n- **Unconditional generation** refers to generating samples unconditionally from the dataset, i.e. $p(y)$\r\n- **[Conditional image generation](/task/conditional-image-generation)** (subtask) refers to generating samples conditionally from the dataset, based on a label, i.e. $p(y|x)$.\r\n\r\nIn this section, you can find state-of-the-art leaderboards for **unconditional generation**. For conditional  generation, and other types of image generations, refer to the subtasks.\r\n\r\n<span style=\"color:grey; opacity: 0.6\">( Image credit: [StyleGAN](https://github.com/NVlabs/stylegan) )</span>","description_from":"task","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","rank":"the archive's row order at snapshot; not re-ranked","rows_end_at":"2025-07-28","rows_withheld_as_spam":0,"metric_values":"the archive's strings, untouched"},"metrics":["FID","bits/dimension","Precision","Recall","FID_CLIP","Model Size (MB)"],"metric_direction":{"note":"inferred from the metric name only (the archive records no direction); null = not inferred, chart draws points only","by_metric":{"FID":"lower","bits/dimension":null,"Precision":"higher","Recall":"higher","FID_CLIP":"lower","Model Size (MB)":null}},"counts":{"rows":39,"rows_with_code":31,"rows_with_paper_page":36,"rows_dated":36,"rows_using_additional_data":0},"rows":[{"rank_in_archive_order":1,"model":"DDPM-IP","metrics":{"FID":"1.27"},"uses_additional_data":false,"paper_date":"2023-01-27","paper":"/paper/input-perturbation-reduces-exposure-bias-in","paper_url":"https://arxiv.org/abs/2301.11706v3","paper_title":"Input Perturbation Reduces Exposure Bias in Diffusion Models","code":"https://github.com/forever208/ddpm-ip","n_code_links":1,"syntology":{"n_ran":2,"n_unverified":1,"n_samples":3,"n_pointer_only_licence":2}},{"rank_in_archive_order":2,"model":"STDDPM-G++","metrics":{"FID":"1.34"},"uses_additional_data":false,"paper_date":"2022-11-28","paper":"/paper/refining-generative-process-with","paper_url":"https://arxiv.org/abs/2211.17091v4","paper_title":"Refining Generative Process with Discriminator Guidance in Score-based Diffusion Models","code":"https://github.com/alsdudrla10/DG","n_code_links":2,"syntology":{"n_ran":2,"n_unverified":0,"n_samples":2,"n_pointer_only_licence":2}},{"rank_in_archive_order":3,"model":"PDM+CS","metrics":{"FID":"1.38"},"uses_additional_data":false,"paper_date":"2023-12-11","paper":"/paper/compensation-sampling-for-improved","paper_url":"https://arxiv.org/abs/2312.06285v1","paper_title":"Compensation Sampling for Improved Convergence in Diffusion Models","code":"https://github.com/hotfinda/Compensation-sampling","n_code_links":1,"syntology":null},{"rank_in_archive_order":4,"model":"Diffusion StyleGAN2","metrics":{"FID":"1.69"},"uses_additional_data":false,"paper_date":"2022-06-05","paper":"/paper/diffusion-gan-training-gans-with-diffusion","paper_url":"https://arxiv.org/abs/2206.02262v4","paper_title":"Diffusion-GAN: Training GANs with Diffusion","code":"https://github.com/Zhendong-Wang/Diffusion-GAN","n_code_links":4,"syntology":{"n_ran":14,"n_unverified":4,"n_samples":18,"n_pointer_only_licence":0}},{"rank_in_archive_order":5,"model":"INDM (VP, FID)","metrics":{"FID":"1.75"},"uses_additional_data":false,"paper_date":"2022-05-27","paper":"/paper/maximum-likelihood-training-of-implicit","paper_url":"https://arxiv.org/abs/2205.13699v3","paper_title":"Maximum Likelihood Training of Implicit Nonlinear Diffusion Models","code":"https://github.com/byeonghu-na/INDM","n_code_links":1,"syntology":{"n_ran":5,"n_unverified":3,"n_samples":8,"n_pointer_only_licence":0}},{"rank_in_archive_order":6,"model":"Patch Diffusion","metrics":{"FID":"1.77"},"uses_additional_data":false,"paper_date":null,"paper":null,"paper_url":null,"paper_title":"","code":null,"n_code_links":0,"syntology":null},{"rank_in_archive_order":7,"model":"LadaGAN","metrics":{"FID":"1.81"},"uses_additional_data":false,"paper_date":"2024-01-17","paper":"/paper/efficient-generative-adversarial-networks","paper_url":"https://arxiv.org/abs/2401.09596v4","paper_title":"Efficient generative adversarial networks using linear additive-attention Transformers","code":"https://github.com/milmor/ladagan","n_code_links":2,"syntology":null},{"rank_in_archive_order":8,"model":"Soft Diffusion (Blur + Noise)","metrics":{"FID":"1.85"},"uses_additional_data":false,"paper_date":"2022-09-12","paper":"/paper/soft-diffusion-score-matching-for-general","paper_url":"https://arxiv.org/abs/2209.05442v2","paper_title":"Soft Diffusion: Score Matching for General Corruptions","code":null,"n_code_links":0,"syntology":null},{"rank_in_archive_order":9,"model":"DDPM++ (VP, FID) + ST","metrics":{"FID":"1.9","bits/dimension":"2.1"},"uses_additional_data":false,"paper_date":"2021-06-10","paper":"/paper/score-matching-model-for-unbounded-data-score-1","paper_url":"https://arxiv.org/abs/2106.05527v5","paper_title":"Soft Truncation: A Universal Training Technique of Score-based Diffusion Model for High Precision Score Estimation","code":"https://github.com/Kim-Dongjun/Soft-Truncation","n_code_links":1,"syntology":{"n_ran":4,"n_unverified":8,"n_samples":12,"n_pointer_only_licence":0}},{"rank_in_archive_order":10,"model":"MMD-PMish-NAS","metrics":{"FID":"1.92","Model Size (MB)":"6.67"},"uses_additional_data":false,"paper_date":"2024-10-23","paper":"/paper/enhancing-gans-with-mmd-neural-architecture","paper_url":"https://ieeexplore.ieee.org/document/10732016","paper_title":"Enhancing GANs with MMD Neural Architecture Search, PMish Activation Function, and Adaptive Rank Decomposition","code":"https://github.com/PrasannaPulakurthi/MMD-PMish-NAS","n_code_links":1,"syntology":null},{"rank_in_archive_order":11,"model":"SciRE-Solver","metrics":{"FID":"2.02"},"uses_additional_data":false,"paper_date":"2023-08-15","paper":"/paper/scire-solver-efficient-sampling-of-diffusion","paper_url":"https://arxiv.org/abs/2308.07896v3","paper_title":"SciRE-Solver: Accelerating Diffusion Models Sampling by Score-integrand Solver with Recursive Difference","code":null,"n_code_links":0,"syntology":null},{"rank_in_archive_order":12,"model":"MMD-PMish-NAS (Compressed)","metrics":{"FID":"2.03","Model Size (MB)":"3.04"},"uses_additional_data":false,"paper_date":"2024-10-23","paper":"/paper/enhancing-gans-with-mmd-neural-architecture","paper_url":"https://ieeexplore.ieee.org/document/10732016","paper_title":"Enhancing GANs with MMD Neural Architecture Search, PMish Activation Function, and Adaptive Rank Decomposition","code":"https://github.com/PrasannaPulakurthi/MMD-PMish-NAS","n_code_links":1,"syntology":null},{"rank_in_archive_order":13,"model":"LEGO","metrics":{"FID":"2.09"},"uses_additional_data":false,"paper_date":"2023-10-10","paper":"/paper/learning-stackable-and-skippable-lego-bricks","paper_url":"https://arxiv.org/abs/2310.06389v3","paper_title":"Learning Stackable and Skippable LEGO Bricks for Efficient, Reconfigurable, and Variable-Resolution Diffusion Modeling","code":"https://github.com/JegZheng/LEGODiffusion","n_code_links":1,"syntology":{"n_ran":13,"n_unverified":4,"n_samples":17,"n_pointer_only_licence":0}},{"rank_in_archive_order":14,"model":"DDIM+CS","metrics":{"FID":"2.11"},"uses_additional_data":false,"paper_date":"2023-12-11","paper":"/paper/compensation-sampling-for-improved","paper_url":"https://arxiv.org/abs/2312.06285v1","paper_title":"Compensation Sampling for Improved Convergence in Diffusion Models","code":"https://github.com/hotfinda/Compensation-sampling","n_code_links":1,"syntology":null},{"rank_in_archive_order":15,"model":"INDM (VE, FID)","metrics":{"FID":"2.54"},"uses_additional_data":false,"paper_date":"2022-05-27","paper":"/paper/maximum-likelihood-training-of-implicit","paper_url":"https://arxiv.org/abs/2205.13699v3","paper_title":"Maximum Likelihood Training of Implicit Nonlinear Diffusion Models","code":"https://github.com/byeonghu-na/INDM","n_code_links":1,"syntology":{"n_ran":5,"n_unverified":3,"n_samples":8,"n_pointer_only_licence":0}},{"rank_in_archive_order":16,"model":"PNDM","metrics":{"FID":"2.71"},"uses_additional_data":false,"paper_date":"2022-02-20","paper":"/paper/pseudo-numerical-methods-for-diffusion-models-1","paper_url":"https://arxiv.org/abs/2202.09778v2","paper_title":"Pseudo Numerical Methods for Diffusion Models on Manifolds","code":"https://github.com/sanster/lama-cleaner","n_code_links":9,"syntology":{"n_ran":10,"n_unverified":4,"n_samples":14,"n_pointer_only_licence":2}},{"rank_in_archive_order":17,"model":"DDPM++ (VP, NLL) + ST","metrics":{"FID":"2.9","bits/dimension":"1.96"},"uses_additional_data":false,"paper_date":"2021-06-10","paper":"/paper/score-matching-model-for-unbounded-data-score-1","paper_url":"https://arxiv.org/abs/2106.05527v5","paper_title":"Soft Truncation: A Universal Training Technique of Score-based Diffusion Model for High Precision Score Estimation","code":"https://github.com/Kim-Dongjun/Soft-Truncation","n_code_links":1,"syntology":{"n_ran":4,"n_unverified":8,"n_samples":12,"n_pointer_only_licence":0}},{"rank_in_archive_order":18,"model":"Blackout Diffusion","metrics":{"FID":"3.22"},"uses_additional_data":false,"paper_date":"2023-05-18","paper":"/paper/blackout-diffusion-generative-diffusion","paper_url":"https://arxiv.org/abs/2305.11089v1","paper_title":"Blackout Diffusion: Generative Diffusion Models in Discrete-State Spaces","code":"https://github.com/lanl/blackout-diffusion","n_code_links":1,"syntology":{"n_ran":2,"n_unverified":4,"n_samples":6,"n_pointer_only_licence":0}},{"rank_in_archive_order":19,"model":"BigGAN-OBRS","metrics":{"FID":"3.74","Precision":"0.74","Recall":"0.65"},"uses_additional_data":false,"paper_date":"2023-11-01","paper":"/paper/optimal-budgeted-rejection-sampling-for","paper_url":"https://arxiv.org/abs/2311.00460v2","paper_title":"Optimal Budgeted Rejection Sampling for Generative Models","code":null,"n_code_links":0,"syntology":null},{"rank_in_archive_order":20,"model":"DiffuseVAE","metrics":{"FID":"3.97"},"uses_additional_data":false,"paper_date":"2022-01-02","paper":"/paper/diffusevae-efficient-controllable-and-high","paper_url":"https://arxiv.org/abs/2201.00308v3","paper_title":"DiffuseVAE: Efficient, Controllable and High-Fidelity Generation from Low-Dimensional Latents","code":"https://github.com/kpandey008/DiffuseVAE","n_code_links":3,"syntology":null},{"rank_in_archive_order":21,"model":"NCP-VAE","metrics":{"FID":"5.25"},"uses_additional_data":false,"paper_date":"2020-10-06","paper":"/paper/ncp-vae-variational-autoencoders-with-noise-1","paper_url":"https://arxiv.org/abs/2010.02917v3","paper_title":"A Contrastive Learning Approach for Training Variational Autoencoder Priors","code":null,"n_code_links":0,"syntology":null},{"rank_in_archive_order":22,"model":"C^3G-NeRF","metrics":{"FID":"5.6"},"uses_additional_data":false,"paper_date":"2023-01-03","paper":"/paper/class-continuous-conditional-generative","paper_url":"https://arxiv.org/abs/2301.00950v3","paper_title":"Class-Continuous Conditional Generative Neural Radiance Field","code":"https://github.com/tom919654/C3G-NeRF","n_code_links":1,"syntology":null},{"rank_in_archive_order":23,"model":"EBM-DRL (T6)","metrics":{"FID":"5.98"},"uses_additional_data":false,"paper_date":"2020-12-15","paper":"/paper/learning-energy-based-models-by-diffusion-1","paper_url":"https://arxiv.org/abs/2012.08125v2","paper_title":"Learning Energy-Based Models by Diffusion Recovery Likelihood","code":"https://github.com/ruiqigao/recovery_likelihood","n_code_links":2,"syntology":null},{"rank_in_archive_order":24,"model":"PR-BigGAN - Recall","metrics":{"FID":"6.03","Precision":"0.78","Recall":"0.56"},"uses_additional_data":false,"paper_date":null,"paper":null,"paper_url":null,"paper_title":"","code":null,"n_code_links":0,"syntology":null},{"rank_in_archive_order":25,"model":"HDCGAN","metrics":{"FID":"8.44"},"uses_additional_data":false,"paper_date":"2017-11-17","paper":"/paper/high-resolution-deep-convolutional-generative","paper_url":"http://arxiv.org/abs/1711.06491v12","paper_title":"High-Resolution Deep Convolutional Generative Adversarial Networks","code":"https://github.com/curto2/c","n_code_links":1,"syntology":null},{"rank_in_archive_order":26,"model":"FDP(UViT)","metrics":{"FID":"11","FID_CLIP":"6.55"},"uses_additional_data":false,"paper_date":"2023-03-01","paper":"/paper/continuous-time-functional-diffusion","paper_url":"https://arxiv.org/abs/2303.00800v3","paper_title":"Continuous-Time Functional Diffusion Processes","code":"https://github.com/giulio98/functional-diffusion-processes","n_code_links":1,"syntology":{"n_ran":3,"n_unverified":0,"n_samples":3,"n_pointer_only_licence":3}},{"rank_in_archive_order":27,"model":"FCE","metrics":{"FID":"12.21"},"uses_additional_data":false,"paper_date":"2019-12-02","paper":"/paper/flow-contrastive-estimation-of-energy-based","paper_url":"https://arxiv.org/abs/1912.00589v2","paper_title":"Flow Contrastive Estimation of Energy-Based Models","code":"https://github.com/ilkhem/icebeem","n_code_links":2,"syntology":{"n_ran":1,"n_unverified":0,"n_samples":1,"n_pointer_only_licence":1}},{"rank_in_archive_order":28,"model":"TransGAN-XL","metrics":{"FID":"12.23"},"uses_additional_data":false,"paper_date":"2021-02-14","paper":"/paper/transgan-two-transformers-can-make-one-strong","paper_url":"https://arxiv.org/abs/2102.07074v4","paper_title":"TransGAN: Two Pure Transformers Can Make One Strong GAN, and That Can Scale Up","code":"https://github.com/VITA-Group/TransGAN","n_code_links":10,"syntology":{"n_ran":8,"n_unverified":0,"n_samples":8,"n_pointer_only_licence":7}},{"rank_in_archive_order":29,"model":"CLR-GAN","metrics":{"FID":"13.63"},"uses_additional_data":false,"paper_date":"2024-09-30","paper":"/paper/clr-gan-improving-gans-stability-and-quality","paper_url":"https://link.springer.com/chapter/10.1007/978-3-031-73232-4_12","paper_title":"CLR-GAN: Improving GANs Stability and Quality via Consistent Latent Representation and Reconstruction","code":"https://github.com/Petecheco/CLR-GAN","n_code_links":1,"syntology":null},{"rank_in_archive_order":30,"model":"PeerGAN","metrics":{"FID":"13.95"},"uses_additional_data":false,"paper_date":"2021-01-19","paper":"/paper/peergan-generative-adversarial-networks-with","paper_url":"https://arxiv.org/abs/2101.07524v3","paper_title":"DuelGAN: A Duel Between Two Discriminators Stabilizes the GAN Training","code":null,"n_code_links":0,"syntology":null},{"rank_in_archive_order":31,"model":"PR-BigGAN - Precision","metrics":{"FID":"22.45","Precision":"0.84","Recall":"0.26"},"uses_additional_data":false,"paper_date":null,"paper":null,"paper_url":null,"paper_title":"","code":null,"n_code_links":0,"syntology":null},{"rank_in_archive_order":32,"model":"BEGAN-CS","metrics":{"FID":"34.136"},"uses_additional_data":false,"paper_date":"2018-08-22","paper":"/paper/escaping-from-collapsing-modes-in-a","paper_url":"http://arxiv.org/abs/1808.07258v1","paper_title":"Escaping from Collapsing Modes in a Constrained Space","code":"https://github.com/chang810249/BEGAN-CS","n_code_links":1,"syntology":null},{"rank_in_archive_order":33,"model":"FDP(MLP)","metrics":{"FID":"35","FID_CLIP":"12.44"},"uses_additional_data":false,"paper_date":"2023-03-01","paper":"/paper/continuous-time-functional-diffusion","paper_url":"https://arxiv.org/abs/2303.00800v3","paper_title":"Continuous-Time Functional Diffusion Processes","code":"https://github.com/giulio98/functional-diffusion-processes","n_code_links":1,"syntology":{"n_ran":3,"n_unverified":0,"n_samples":3,"n_pointer_only_licence":3}},{"rank_in_archive_order":34,"model":"PAE","metrics":{"FID":"49.2"},"uses_additional_data":false,"paper_date":"2020-06-09","paper":"/paper/probabilistic-auto-encoder","paper_url":"https://arxiv.org/abs/2006.05479v4","paper_title":"Probabilistic Autoencoder","code":"https://github.com/VMBoehm/PAE","n_code_links":4,"syntology":{"n_ran":1,"n_unverified":12,"n_samples":13,"n_pointer_only_licence":0}},{"rank_in_archive_order":35,"model":"Feature Alignment","metrics":{"FID":"128.35"},"uses_additional_data":false,"paper_date":"2021-06-23","paper":"/paper/feature-alignment-for-approximated","paper_url":"https://arxiv.org/abs/2106.12562v2","paper_title":"Feature Alignment as a Generative Process","code":"https://github.com/tiago939/feature_alignment","n_code_links":2,"syntology":null},{"rank_in_archive_order":36,"model":"Efficient-VDVAE","metrics":{"bits/dimension":"1.83"},"uses_additional_data":false,"paper_date":"2022-03-25","paper":"/paper/efficient-vdvae-less-is-more","paper_url":"https://arxiv.org/abs/2203.13751v2","paper_title":"Efficient-VDVAE: Less is more","code":"https://github.com/Rayhane-mamah/Efficient-VDVAE","n_code_links":1,"syntology":null},{"rank_in_archive_order":37,"model":"CR-NVAE","metrics":{"bits/dimension":"1.86"},"uses_additional_data":false,"paper_date":"2021-05-31","paper":"/paper/consistency-regularization-for-variational","paper_url":"https://arxiv.org/abs/2105.14859v2","paper_title":"Consistency Regularization for Variational Auto-Encoders","code":"https://github.com/sinhasam/crvae","n_code_links":1,"syntology":{"n_ran":0,"n_unverified":3,"n_samples":3,"n_pointer_only_licence":0}},{"rank_in_archive_order":38,"model":"UNCSN++ (RVE) + ST","metrics":{"bits/dimension":"1.97"},"uses_additional_data":false,"paper_date":"2021-06-10","paper":"/paper/score-matching-model-for-unbounded-data-score-1","paper_url":"https://arxiv.org/abs/2106.05527v5","paper_title":"Soft Truncation: A Universal Training Technique of Score-based Diffusion Model for High Precision Score Estimation","code":"https://github.com/Kim-Dongjun/Soft-Truncation","n_code_links":1,"syntology":{"n_ran":4,"n_unverified":8,"n_samples":12,"n_pointer_only_licence":0}},{"rank_in_archive_order":39,"model":"DenseFlow-74-10","metrics":{"bits/dimension":"1.99"},"uses_additional_data":false,"paper_date":"2021-06-08","paper":"/paper/densely-connected-normalizing-flows","paper_url":"https://arxiv.org/abs/2106.04627v3","paper_title":"Densely connected normalizing flows","code":"https://github.com/matejgrcic/DenseFlow","n_code_links":4,"syntology":null}],"since_archive":{"claim":"Results that newer papers report for their own method, placed here by Syntology. A model pointed at the cell in the paper's own table; the number was read from that cell and checked against this leaderboard's metric, dataset, split and scale; an independent check that saw this leaderboard's other rows and every other leaderboard on the same dataset accepted it. Not reviewed by the paper's authors or by the archive's editors, and not ranked against the archive rows.","extraction_file_present":true,"measurement":{"test_papers":883,"papers_with_output":881,"judged_true":108,"judged":110,"wilson95_lower":0.9361,"measured_on":"2026-09-24","frozen_commit":"0e3de0df94"},"measurement_note":"blind adjudication of accepted entries on a held-out split of archive papers, rules frozen before the test","coverage":{"sentence":"Syntology has checked 6,885 of the 9,623 papers on this site that are newer than the archive; results from the others appear after they are checked.","complete":false,"papers_newer_than_archive":9623,"papers_checked":6885,"papers_extracted_not_yet_verified":0,"boards_without_verdict":2,"papers_not_yet_extracted":2737},"order":"newest first by month (arXiv date, else the arXiv-id month), then arXiv id descending","columns":[],"entries":[]},"syntology":{"read_at":"2026-09-25T09:33:49+00:00","claim":"Per row: N of M harvested code samples from that row's paper executed on a synthesized fixture; the other M-N are unverified. Not a reproduction of the row's number; not a correctness claim. n_pointer_only_licence counts samples the site points at rather than redistributes (a licence axis, independent of ran/unverified).","rows_with_graph_line":17,"rows_with_any_sample_ran":16,"distinct_papers_with_graph_line":13,"distinct_papers_with_any_sample_ran":12,"samples_over_distinct_papers":{"n_ran":65,"n_unverified":43,"n_samples":108,"n_pointer_only_licence":17,"note":"each paper (arXiv id) counted once, however many rows it is behind; this is the page-level figure"},"samples_row_weighted":{"n_ran":81,"n_unverified":62,"n_samples":143,"n_pointer_only_licence":20,"note":"row-weighted: a paper behind several rows is counted once per row; inflated relative to samples_over_distinct_papers by design, kept for readers summing the per-row syntology blocks"}}}