AI Paper Atlas

Research map / Benchmarks / CIFAR-10

CIFAR-10 leaderboard

1,996 papers at ICML, NeurIPS, ICLR, CVPR, AAAI, ICCV and ECCV, and the journals TPAMI, IJCV and JMLR report results on CIFAR-10. Best accuracy: 99.94 (S2MVTC, CVPR 2024).

Numbers are as each paper reports them, read by AI from its results tables and not checked by hand. Papers can differ in setting (training data, model size, extra data); the topic column shows what each paper is about. Open the paper before you compare or quote a number.

CIFAR-10 · accuracy

1,135 papers, each with its best reported number for its own method; higher is better.

1,125 more papers on this leaderboard, the baselines reported on it and a CSV download are free with a scime account.
Sign in free to see all

CIFAR-10 · error rate

277 papers, each with its best reported number for its own method; lower is better.

#error rate ↓MethodPaperTopic
11.2IHDA+
PyramidNet
Post-training Iterative Hierarchical Data Augmentation for Deep Networks
NeurIPS 2020 · Table 1
Recognition and long-tail
21.28Few-shot DARTS-Large (cutout + autoaug)
DARTS
Few-Shot Neural Architecture Search
ICML 2021 · Table: CIFAR-10 NASNet search
Neural architecture search
31.46PBA
PyramidNet+ShakeDrop
Population Based Augmentation: Efficient Learning of Augmentation Policy Schedules
ICML 2019 · Table 2
Visual and representation RL
41.47MC-CP
CNN
Robust Uncertainty Quantification Using Conformalised Monte Carlo Prediction
AAAI 2024 · Table 1
Uncertainty and Gaussian processes
51.5PyramidNet+ShakeDrop + AutoAugment
PyramidNet+ShakeDrop
AutoAugment: Learning Augmentation Strategies From Data
CVPR 2019 · Table 2
Neural architecture search
61.7Proposed
PyramidNet+ShakeDrop
Learning Augmentation Network via Influence Functions
CVPR 2020 · Table 2
Recognition and long-tail
71.7DADA
PyramidNet+ShakeDrop
DADA: Differentiable Automatic Data Augmentation
ECCV 2020 · Table 2
Neural architecture search
81.8CutMix + SLA+SD
PyramidNet200
Self-supervised Label Augmentation via Input Transformations
ICML 2020 · Table 4
Self-supervised and contrastive
91.89OHL-Auto-Aug
AmoebaNet-B(6,128)
Online Hyper-Parameter Learning for Auto-Augmentation Strategy
ICCV 2019 · Table 2
Neural architecture search
101.93NAONet + CutoutNeural Architecture Optimization
NeurIPS 2018 · Table 1
Neural architecture search
267 more papers on this leaderboard, the baselines reported on it and a CSV download are free with a scime account.
Sign in free to see all

CIFAR-10 · FID

263 papers, each with its best reported number for its own method; lower is better.

#FID ↓MethodPaperTopic
11.74DDDM (T=8000, 1000 NFE)Directly Denoising Diffusion Models
ICML 2024 · CIFAR-10 results table
Fast diffusion sampling
21.77EDM-G++ with BCE density-ratio matching
EDM
Refining Generative Process with Discriminator Guidance in Score-based Diffusion Models
ICML 2023 · Table: Ablation study for h-fu
Fast diffusion sampling
31.81RINScalable Adaptive Computation for Iterative Generation
ICML 2023 · text
Autoregressive and latent generation
41.88VP (unfrozen) + LEGO-PR
VP
Learning Stackable and Skippable LEGO Bricks for Efficient, Reconfigurable, and Variable-Resolution Diffusion Modeling
ICLR 2024 · Table: CIFAR10 pretrained mode
Autoregressive and latent generation
51.9STF - Heun, NCSN++
EDM, NCSN++
Stable Target Field for Reduced Variance Score Estimation in Diffusion Models
ICLR 2023 · Image generation table
Fast diffusion sampling
61.93Residual correction + EDM
EDM
Residual Learning in Diffusion Models
CVPR 2024 · Table 2
Image restoration
71.96Multistage EDM
EDM
Improving Training Efficiency of Diffusion Models via Multi-Stage Framework and Tailored Multi-Decoder Architecture
CVPR 2024 · Table 1
Fast diffusion sampling
81.98Simple ReFlow + GuidanceSimple ReFlow: Improved Techniques for Fast Flow Models
ICLR 2025 · Comparison of neural ODE metho
Fast diffusion sampling
91.98IMM, 2 stepsInductive Moment Matching
ICML 2025 · Table 1
Fast diffusion sampling
101.99Corrector optimizedScore-Optimal Diffusion Schedules
NeurIPS 2024 · Table 1
Fast diffusion sampling
253 more papers on this leaderboard, the baselines reported on it and a CSV download are free with a scime account.
Sign in free to see all

CIFAR-10 · Inception score

60 papers, each with its best reported number for its own method; higher is better.

50 more papers on this leaderboard, the baselines reported on it and a CSV download are free with a scime account.
Sign in free to see all

CIFAR-10 · IS

54 papers, each with its best reported number for its own method; higher is better.

44 more papers on this leaderboard, the baselines reported on it and a CSV download are free with a scime account.
Sign in free to see all

CIFAR-10 · clean accuracy

51 papers, each with its best reported number for its own method; higher is better.

#clean accuracy ↑MethodPaperTopic
197.07Co-training, radius 2/255
ViT-S
Revisiting adapters with adversarial training
ICLR 2023 · Table: Evaluating on other dat
Certified and empirical robustness
296.1PubDef
WideResNet-34-10
PubDef: Defending Against Transfer Attacks From Public Models
ICLR 2024 · Table, Clean and adversarial a
Black-box and transfer attacks
395.8VIPAug-F
ResNet-18
Domain Generalization with Vital Phase Augmentation
AAAI 2024 · CIFAR-10 and CIFAR-100 compari
Domain adaptation and generalization
495.04SL
ResNet-18
Understanding Square Loss in Training Overparametrized Neural Network Classifiers
NeurIPS 2022 · Table 1
Kernels and regression theory
594.9TAC (L1)
WideResNet-28-10
Constraining Representations Yields Models That Know What They Don't Know
ICLR 2023 · Appendix: Further results on t
Recognition and long-tail
694.4Noisy mixup
ResNet-18
Boundary thickness and robustness in learning models
NeurIPS 2020 · Table 1
Certified and empirical robustness
794.38SPD
ResNet-18
SPD: Shallow Backdoor Protecting Deep Backdoor Against Backdoor Detection
ICCV 2025 · Table 1
Backdoor and poisoning
894.18RoLI - Full-FT
Swin Transformer
Initialization Matters for Adversarial Transfer Learning
CVPR 2024 · Table 2
Certified and empirical robustness
993.82PRL
ResNet-18
Probabilistically Robust Learning: Balancing Average and Worst-case Performance
ICML 2022 · Table 1
Certified and empirical robustness
1093.8ℓ∞ norm decay ResNet-18, β=10^-3
ResNet-18
Large Norms of CNN Layers Do Not Hurt Adversarial Robustness
AAAI 2021 · Table: norm-regularization com
Certified and empirical robustness
41 more papers on this leaderboard, the baselines reported on it and a CSV download are free with a scime account.
Sign in free to see all

CIFAR-10 · AUROC

38 papers, each with its best reported number for its own method; higher is better.

28 more papers on this leaderboard, the baselines reported on it and a CSV download are free with a scime account.
Sign in free to see all

CIFAR-10 · top-1 error

35 papers, each with its best reported number for its own method; lower is better.

#top-1 error ↓MethodPaperTopic
11.29MCMC-Aug
PyramidNet
A Continuous Mapping For Augmentation Design
NeurIPS 2021 · Table 3
Optimizers
22OnlineAugment + AutoAugment
Wide-ResNet-28-10
OnlineAugment: Online Data Augmentation with Less Domain Knowledge
ECCV 2020 · Table 6
Domain adaptation and generalization
32.35RM
PyramidNet-200
RecursiveMix: Mixed Learning with History
NeurIPS 2022 · Table 1
CNN features
42.55Pyramid-200 + StyleCutMix (auto-γ)
PyramidNet-200
StyleMix: Separating Content and Style for Enhanced Data Augmentation
CVPR 2021 · Table 4
Certified and empirical robustness
52.62EcoNASEcoNAS: Finding Proxies for Economical Neural Architecture Search
CVPR 2020 · CIFAR-10 results table
Neural architecture search
62.76WideResNet-28-10 + SaliencyMix
WideResNet-28-10
SaliencyMix: A Saliency Guided Data Augmentation Strategy for Better Regularization
ICLR 2021 · Table 1
CNN features
72.88CutMix
PyramidNet-200
CutMix: Regularization Strategy to Train Strong Classifiers With Localizable Features
ICCV 2019 · Table 6
Weakly supervised detection and segmentation
83.08Random Erasing
WRN-28-10
Random Erasing Data Augmentation
AAAI 2020 · Table 1
CNN features
93.31PyramidNet bottleneck, α=200Deep Pyramidal Residual Networks
CVPR 2017 · Table: CIFAR results
CNN design
103.33SGD + DCL-inf-1
ResNeXt-29
Direction Concentration Learning: Enhancing Congruency in Machine Learning
TPAMI 2019 · Table 6
Online and replay-based
25 more papers on this leaderboard, the baselines reported on it and a CSV download are free with a scime account.
Sign in free to see all

CIFAR-10 · ECE

27 papers, each with its best reported number for its own method; lower is better.

#ECE ↓MethodPaperTopic
10.077Mixup (α=1.0)
WideResNet-40-10
Pushing Boundaries: Mixup's Influence on Neural Collapse
ICLR 2024 · CIFAR10 expected calibration e
Recognition and long-tail
20.088GSD ResNet34
ResNet34
A Geometric Perspective towards Neural Calibration via Sensitivity Decomposition
NeurIPS 2021 · Table, Generalizability Experi
Recognition and long-tail
30.29QaTS
DN-121
Quantile‑Adaptive Temperature Scaling for Confidence Calibration
ECCV 2026 · Table 1
Recognition and long-tail
40.41LogitNorm with temperature scaling
WRN-40-2
Mitigating Neural Network Overconfidence with Logit Normalization
ICML 2022 · Table 5
OOD and anomaly detection
50.45AugCAMixup BE
Wide ResNet-28-10 BatchEnsemble
Combining Ensembles and Data Augmentation Can Harm Your Calibration
ICLR 2021 · AugMix BatchEnsemble table
Recognition and long-tail
60.66FED
ResNet18
Functional Ensemble Distillation
NeurIPS 2022 · Table: Test accuracy, agreemen
Uncertainty and Gaussian processes
70.76SMART
ResNet-50
Sample Margin-Aware Recalibration of Temperature Scaling
ICML 2026 · Table 1
Prediction and decision losses
80.76BalCAL
WRN-28-10
Balancing Two Classifiers via A Simplex ETF Structure for Model Calibration
CVPR 2025 · Table 1
Recognition and long-tail
90.76Adaptive TS (CE)
WideResNet28-10
Sample-Dependent Adaptive Temperature Scaling for Improved Calibration
AAAI 2023 · Calibration results table
RNNs and normalization
100.84PTS
VGG19
Parameterized Temperature Scaling for Boosting the Expressive Power in Post-Hoc Uncertainty Calibration
ECCV 2022 · Table 1
Recognition and long-tail
17 more papers on this leaderboard, the baselines reported on it and a CSV download are free with a scime account.
Sign in free to see all

CIFAR-10 · NLL

25 papers, each with its best reported number for its own method; lower is better.

15 more papers on this leaderboard, the baselines reported on it and a CSV download are free with a scime account.
Sign in free to see all

CIFAR-10 · accuracy (val)

24 papers, each with its best reported number for its own method; higher is better.

#accuracy ↑MethodPaperTopic
195.94BatchEnsemble
ResNet32x4
BatchEnsemble: an Alternative Approach to Efficient Ensemble and Lifelong Learning
ICLR 2020 · Classification table
RNNs and normalization
295.6Ours (VR-PGE sparse training)
WRN-28-10
Efficient Neural Network Training via Forward and Backward Propagation Sparsification
NeurIPS 2021 · Table 2
Optimizers
395.4ACLS
ResNet-50
ACLS: Adaptive and Conditional Label Smoothing for Network Calibration
ICCV 2023 · Table 3
Recognition and long-tail
495WRN-40-4 + XD-operations
WRN-40-4
Rethinking Neural Operations for Diverse Tasks
NeurIPS 2021 · Table (CIFAR-10 search space c
Neural architecture search
594.41Coded ResNeXt-29 (10x11d)Towards Disentangling Information Paths with Coded ResNeXt
NeurIPS 2022 · Table (default hyperparameters
RNNs and normalization
69416-CNN ensembleDo Deep Convolutional Nets Really Need to be Deep and Convolutional?
ICLR 2017 · text
CNN design
793.97LGL(r, 4)
VGG-16
Local to Global Learning: Gradually Adding Classes for Training Deep Neural Networks
CVPR 2019 · Table 1
Recognition and long-tail
893.9MandelbulbVAR-1k
ResNet-50
Pre-training Vision Models with Mandelbulb Variations
CVPR 2024 · Table 1
Attention and transformers
993.43Winograd-ReLU CNN
VGG-nagadomi
Efficient Sparse-Winograd Convolutional Neural Networks
ICLR 2018 · text
CNN design
1093.2S2FP8
ResNet-50
Shifted and Squeezed 8-bit Floating Point format for Low-Precision Training of Deep Neural Networks
ICLR 2020 · Table: CIFAR-10 validation acc
Quantization
14 more papers on this leaderboard, the baselines reported on it and a CSV download are free with a scime account.
Sign in free to see all

CIFAR-10 · natural accuracy

23 papers, each with its best reported number for its own method; higher is better.

#natural accuracy ↑MethodPaperTopic
195.5SAM (ρ=0.2)
PreActResNet-18
On the Duality Between Sharpness-Aware Minimization and Adversarial Training
ICML 2024 · Table: Natural and robust accu
Certified and empirical robustness
294.56Natural training+CASImproving Adversarial Robustness via Channel-wise Activation Suppressing
ICLR 2021 · CAS Improves Representation Le
Certified and empirical robustness
391.3WsrNet-Plus (TRADES-1)Neural Architecture Search for Wide Spectrum Adversarial Robustness
AAAI 2023 · Table 1
Certified and empirical robustness
491.03Generalist
WRN-32-10
Generalist: Decoupling Natural and Robust Generalization
CVPR 2023 · Table 1(b)
Certified and empirical robustness
590.58DCS + GSAT
WRN 34
Adversarial Robustness via Deformable Convolution with Stochasticity
ICML 2025 · Table 1
Certified and empirical robustness
690.01RST+MeoW
WideResNet-28-10
Memorization Weights for Instance Reweighting in Adversarial Training
AAAI 2023 · Table 3
Certified and empirical robustness
789.94FAT for TRADES (εtrain=8/255, β=1.0)
WRN-34-10
Attacks Which Do Not Kill Training Make Adversarial Learning Stronger
ICML 2020 · Table 2
Certified and empirical robustness
889.35VITA for Adv. Training
WRN-34-10
VITA: A Multi-Source Vicinal Transfer Augmentation Method for Out-of-Distribution Generalization
AAAI 2022 · Table 3
Recognition and long-tail
986.61NADAR-A
WRN32-10
Neural Architecture Dilation for Adversarial Robustness
NeurIPS 2021 · Table: CIFAR-10 validation acc
Certified and empirical robustness
1086.6AT+CR
WideResNet34
A Unified Contrastive Energy-based Model for Understanding the Generative Ability of Adversarial Training
ICLR 2022 · Appendix: supervised robustnes
Certified and empirical robustness
13 more papers on this leaderboard, the baselines reported on it and a CSV download are free with a scime account.
Sign in free to see all

18 less reported leaderboards for CIFAR-10 (other metrics or splits) are in the CSV and the API.

Not affiliated with any conference, journal or the benchmark's authors. Corrections: contact@latenticlabs.com.