# Image Classification on CIFAR10 (Clean and PGD Robustness)

Source: [Image Classification on CIFAR10 (Clean and PGD Robustness)](<https://www.sota2.com/research/sota/image-classification-on-cifar10-clean-and-pgd-robustness>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-02-27T11:19:52.273Z

Evaluation of Image Classification performance on the CIFAR-10 dataset, reporting standard clean accuracy alongside robustness metrics derived from Projected Gradient Descent (PGD) attacks.

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

| Scope | Recorded value |
| --- | --- |
| Task | Image Classification |
| Dataset (including any stated split) | CIFAR10 |
| Metric | Clean Accuracy |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 37 of 37 |
| Coverage | All results currently returned for this benchmark |

Evaluation settings may differ between results. The original papers and recorded configurations are needed to establish comparability. A shared metric name alone does not establish an identical protocol.

## Metrics

- [Clean Accuracy](<https://www.sota2.com/research/sota/image-classification-on-cifar10-clean-and-pgd-robustness.md?metric=Clean%20Accuracy>): Higher is better; unit: PERCENT.

- [PGD Accuracy](<https://www.sota2.com/research/sota/image-classification-on-cifar10-clean-and-pgd-robustness.md?metric=PGD%20Accuracy>): Higher is better; unit: PERCENT.

## Recorded results

| Method | Clean Accuracy | Result date | Report paper |
| --- | --- | --- | --- |
| LPF | 94.3 | Not recorded | 69799806a2b28e761d87cb52 |
| RoLI - Full-FT | 94.18 | Not recorded | 68fa72bf1e209131cea585bb |
| TTC | 94.1 | Not recorded | 69799806a2b28e761d87cb52 |
| CLIP | 93.4 | Not recorded | 69799806a2b28e761d87cb52 |
| RanLI - Full-FT | 93.02 | Not recorded | 68fa72bf1e209131cea585bb |
| TTE | 92.3 | Not recorded | 69799806a2b28e761d87cb52 |
| RoLI - LoRA | 92.23 | Not recorded | 68fa72bf1e209131cea585bb |
| HD | 92.1 | Not recorded | 69799806a2b28e761d87cb52 |
| RoLI - Bias | 91.83 | Not recorded | 68fa72bf1e209131cea585bb |
| TWINS-AT | 91.24 | Not recorded | 68fa72bf1e209131cea585bb |
| HPT-GPD | 90.74 | Not recorded | 69705d96e25ad2fa9d1ed1a8 |
| CSR | 90.7 | Not recorded | 69799806a2b28e761d87cb52 |
| RoLI - Adapter | 90.31 | Not recorded | 68fa72bf1e209131cea585bb |
| RanLI - Bias | 90.25 | Not recorded | 68fa72bf1e209131cea585bb |
| RanLI - VPT | 90.24 | Not recorded | 68fa72bf1e209131cea585bb |
| RanLI - Adapter | 90.23 | Not recorded | 68fa72bf1e209131cea585bb |
| R-TPT | 90.2 | Not recorded | 69799806a2b28e761d87cb52 |
| RanLI - LoRA | 90.02 | Not recorded | 68fa72bf1e209131cea585bb |
| Anti-Adv | 89.3 | Not recorded | 69799806a2b28e761d87cb52 |
| FT-Standard | 88.67 | Not recorded | 69705d96e25ad2fa9d1ed1a8 |
| TTC | 87.6 | Not recorded | 69799806a2b28e761d87cb52 |
| CLIP | 87.43 | Not recorded | 69705d96e25ad2fa9d1ed1a8 |
| AutoLoRa | 86.93 | Not recorded | 68fa72bf1e209131cea585bb |
| HD | 86.5 | Not recorded | 69799806a2b28e761d87cb52 |
| CLIP | 86.1 | Not recorded | 69799806a2b28e761d87cb52 |
| CSR | 86.1 | Not recorded | 69799806a2b28e761d87cb52 |
| TTE | 86 | Not recorded | 69799806a2b28e761d87cb52 |
| LPF | 84.9 | Not recorded | 69799806a2b28e761d87cb52 |
| Anti-Adv | 84.4 | Not recorded | 69799806a2b28e761d87cb52 |
| RanLI - Linear | 84.16 | Not recorded | 68fa72bf1e209131cea585bb |
| PMG | 82.8 | Not recorded | 69705d96e25ad2fa9d1ed1a8 |
| R-TPT | 76.7 | Not recorded | 69799806a2b28e761d87cb52 |
| FARE | 76.6 | Not recorded | 69799806a2b28e761d87cb52 |
| TeCoA | 73.7 | Not recorded | 69799806a2b28e761d87cb52 |
| FT-TeCoA | 66.79 | Not recorded | 69705d96e25ad2fa9d1ed1a8 |
| TeCoA | 50.3 | Not recorded | 69799806a2b28e761d87cb52 |
| FARE | 47.2 | Not recorded | 69799806a2b28e761d87cb52 |

### LPF

Clean Accuracy: 94.3. Result ID: 69a17d1735dfcabb00e76284.

[Reporting paper](<https://www.sota2.com/research/paper/69799806a2b28e761d87cb52.md>)

Recorded evaluation configuration: {"Backbone":"CLIP-L/14","Method category":"Test-Time Defense"}. This record does not assert independent reproduction.

### RoLI - Full-FT

Clean Accuracy: 94.18. Result ID: 6992775a3a6c3148490f278d.

[Reporting paper](<https://www.sota2.com/research/paper/68fa72bf1e209131cea585bb.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa72bf1e209131cea585bb.md>)

[Implementation](<https://github.com/DongXzz/RoLI>)

Recorded evaluation configuration: {"Backbone":"Swin Transformer","Initialization":"Robust Linear Initialization","Finetuning Protocol":"Full Finetuning"}. This record does not assert independent reproduction.

### TTC

Clean Accuracy: 94.1. Result ID: 69a17d1735dfcabb00e76288.

[Reporting paper](<https://www.sota2.com/research/paper/69799806a2b28e761d87cb52.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa774a1e209131cea724cc.md>)

[Implementation](<https://www.sota2.com/here>)

Recorded evaluation configuration: {"Backbone":"CLIP-L/14","Method category":"Test-Time Defense"}. This record does not assert independent reproduction.

### CLIP

Clean Accuracy: 93.4. Result ID: 69a17d1735dfcabb00e76280.

[Reporting paper](<https://www.sota2.com/research/paper/69799806a2b28e761d87cb52.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa69581e209131cea21ad0.md>)

[Implementation](<https://github.com/OpenAI/CLIP>)

Recorded evaluation configuration: {"Backbone":"CLIP-L/14","Method category":"Original"}. This record does not assert independent reproduction.

### RanLI - Full-FT

Clean Accuracy: 93.02. Result ID: 6992775a3a6c3148490f2787.

[Reporting paper](<https://www.sota2.com/research/paper/68fa72bf1e209131cea585bb.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa671d1e209131cea145d5.md>)

Recorded evaluation configuration: {"Backbone":"Swin Transformer","Initialization":"Random Initialization","Finetuning Protocol":"Full Finetuning"}. This record does not assert independent reproduction.

### TTE

Clean Accuracy: 92.3. Result ID: 69a17d1735dfcabb00e76287.

[Reporting paper](<https://www.sota2.com/research/paper/69799806a2b28e761d87cb52.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6a711e209131cea2817e.md>)

Recorded evaluation configuration: {"Backbone":"CLIP-L/14","Method category":"Test-Time Defense"}. This record does not assert independent reproduction.

### RoLI - LoRA

Clean Accuracy: 92.23. Result ID: 6992775a3a6c3148490f278f.

[Reporting paper](<https://www.sota2.com/research/paper/68fa72bf1e209131cea585bb.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa72bf1e209131cea585bb.md>)

[Implementation](<https://github.com/DongXzz/RoLI>)

Recorded evaluation configuration: {"Backbone":"Swin Transformer","Initialization":"Robust Linear Initialization","Finetuning Protocol":"LoRA"}. This record does not assert independent reproduction.

### HD

Clean Accuracy: 92.1. Result ID: 69a17d1735dfcabb00e76285.

[Reporting paper](<https://www.sota2.com/research/paper/69799806a2b28e761d87cb52.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6a1d1e209131cea26045.md>)

Recorded evaluation configuration: {"Backbone":"CLIP-L/14","Method category":"Test-Time Defense"}. This record does not assert independent reproduction.

### RoLI - Bias

Clean Accuracy: 91.83. Result ID: 6992775a3a6c3148490f2790.

[Reporting paper](<https://www.sota2.com/research/paper/68fa72bf1e209131cea585bb.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa72bf1e209131cea585bb.md>)

[Implementation](<https://github.com/DongXzz/RoLI>)

Recorded evaluation configuration: {"Backbone":"Swin Transformer","Initialization":"Robust Linear Initialization","Finetuning Protocol":"Bias-only"}. This record does not assert independent reproduction.

### TWINS-AT

Clean Accuracy: 91.24. Result ID: 6992775a3a6c3148490f2785.

[Reporting paper](<https://www.sota2.com/research/paper/68fa72bf1e209131cea585bb.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6ed11e209131cea4265b.md>)

[Implementation](<https://github.com/ziquanliu/CVPR2023-TWINS>)

Recorded evaluation configuration: {"Backbone":"ResNet-50"}. This record does not assert independent reproduction.

### HPT-GPD

Clean Accuracy: 90.74. Result ID: 69a17dcc35dfcabb00e7657d.

[Reporting paper](<https://www.sota2.com/research/paper/69705d96e25ad2fa9d1ed1a8.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69705d96e25ad2fa9d1ed1a8.md>)

Recorded evaluation configuration: {"Backbone":"ViT-B/32","Proxy CLIP":"ViT-B/16","Attack Protocol":"PGD-10","Training time per epoch (s)":"1486"}. This record does not assert independent reproduction.

### CSR

Clean Accuracy: 90.7. Result ID: 69a17d1735dfcabb00e76289.

[Reporting paper](<https://www.sota2.com/research/paper/69799806a2b28e761d87cb52.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69799806a2b28e761d87cb52.md>)

Recorded evaluation configuration: {"Backbone":"CLIP-L/14","Method category":"Test-Time Defense"}. This record does not assert independent reproduction.

### RoLI - Adapter

Clean Accuracy: 90.31. Result ID: 6992775a3a6c3148490f278e.

[Reporting paper](<https://www.sota2.com/research/paper/68fa72bf1e209131cea585bb.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa72bf1e209131cea585bb.md>)

[Implementation](<https://github.com/DongXzz/RoLI>)

Recorded evaluation configuration: {"Backbone":"Swin Transformer","Initialization":"Robust Linear Initialization","Finetuning Protocol":"Adapter"}. This record does not assert independent reproduction.

### RanLI - Bias

Clean Accuracy: 90.25. Result ID: 6992775a3a6c3148490f278a.

[Reporting paper](<https://www.sota2.com/research/paper/68fa72bf1e209131cea585bb.md>)

Recorded evaluation configuration: {"Backbone":"Swin Transformer","Initialization":"Random Initialization","Finetuning Protocol":"Bias-only"}. This record does not assert independent reproduction.

### RanLI - VPT

Clean Accuracy: 90.24. Result ID: 6992775a3a6c3148490f278b.

[Reporting paper](<https://www.sota2.com/research/paper/68fa72bf1e209131cea585bb.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6d181e209131cea37ae9.md>)

[Implementation](<https://www.sota2.com/github.com/kmnp/vpt>)

Recorded evaluation configuration: {"Backbone":"Swin Transformer","Initialization":"Random Initialization","Finetuning Protocol":"Visual Prompt Tuning"}. This record does not assert independent reproduction.

### RanLI - Adapter

Clean Accuracy: 90.23. Result ID: 6992775a3a6c3148490f2788.

[Reporting paper](<https://www.sota2.com/research/paper/68fa72bf1e209131cea585bb.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa65e31e209131cea0d324.md>)

[Implementation](<https://github.com/google-research/adapter-bert>)

Recorded evaluation configuration: {"Backbone":"Swin Transformer","Initialization":"Random Initialization","Finetuning Protocol":"Adapter"}. This record does not assert independent reproduction.

### R-TPT

Clean Accuracy: 90.2. Result ID: 69a17d1735dfcabb00e76283.

[Reporting paper](<https://www.sota2.com/research/paper/69799806a2b28e761d87cb52.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7add1e209131cea8590d.md>)

[Implementation](<https://github.com/TomSheng21/R-TPT>)

Recorded evaluation configuration: {"Backbone":"CLIP-L/14","Method category":"Test-Time Defense"}. This record does not assert independent reproduction.

### RanLI - LoRA

Clean Accuracy: 90.02. Result ID: 6992775a3a6c3148490f2789.

[Reporting paper](<https://www.sota2.com/research/paper/68fa72bf1e209131cea585bb.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6b121e209131cea2bf59.md>)

[Implementation](<https://github.com/microsoft/LoRA>)

Recorded evaluation configuration: {"Backbone":"Swin Transformer","Initialization":"Random Initialization","Finetuning Protocol":"LoRA"}. This record does not assert independent reproduction.

### Anti-Adv

Clean Accuracy: 89.3. Result ID: 69a17d1735dfcabb00e76286.

[Reporting paper](<https://www.sota2.com/research/paper/69799806a2b28e761d87cb52.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6b7f1e209131cea2e9af.md>)

Recorded evaluation configuration: {"Backbone":"CLIP-L/14","Method category":"Test-Time Defense"}. This record does not assert independent reproduction.

### FT-Standard

Clean Accuracy: 88.67. Result ID: 69a17dcc35dfcabb00e7657a.

[Reporting paper](<https://www.sota2.com/research/paper/69705d96e25ad2fa9d1ed1a8.md>)

Recorded evaluation configuration: {"Backbone":"ViT-B/32","Proxy CLIP":"ViT-B/16","Attack Protocol":"PGD-10","Training time per epoch (s)":"406"}. This record does not assert independent reproduction.

### TTC

Clean Accuracy: 87.6. Result ID: 69a17d1735dfcabb00e7612b.

[Reporting paper](<https://www.sota2.com/research/paper/69799806a2b28e761d87cb52.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa774a1e209131cea724cc.md>)

[Implementation](<https://www.sota2.com/here>)

Recorded evaluation configuration: {"Backbone":"CLIP-B/32","Category":"Test-Time Defense"}. This record does not assert independent reproduction.

### CLIP

Clean Accuracy: 87.43. Result ID: 69a17dcc35dfcabb00e76579.

[Reporting paper](<https://www.sota2.com/research/paper/69705d96e25ad2fa9d1ed1a8.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa69581e209131cea21ad0.md>)

[Implementation](<https://github.com/OpenAI/CLIP>)

Recorded evaluation configuration: {"Backbone":"ViT-B/32","Proxy CLIP":"ViT-B/16","Attack Protocol":"PGD-10","Training time per epoch (s)":"0"}. This record does not assert independent reproduction.

### AutoLoRa

Clean Accuracy: 86.93. Result ID: 6992775a3a6c3148490f2786.

[Reporting paper](<https://www.sota2.com/research/paper/68fa72bf1e209131cea585bb.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa70b91e209131cea4cd55.md>)

Recorded evaluation configuration: {"Backbone":"ResNet-50"}. This record does not assert independent reproduction.

### HD

Clean Accuracy: 86.5. Result ID: 69a17d1735dfcabb00e76128.

[Reporting paper](<https://www.sota2.com/research/paper/69799806a2b28e761d87cb52.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6a1d1e209131cea26045.md>)

Recorded evaluation configuration: {"Backbone":"CLIP-B/32","Category":"Test-Time Defense"}. This record does not assert independent reproduction.

### CLIP

Clean Accuracy: 86.1. Result ID: 69a17d1735dfcabb00e76123.

[Reporting paper](<https://www.sota2.com/research/paper/69799806a2b28e761d87cb52.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa69581e209131cea21ad0.md>)

[Implementation](<https://github.com/OpenAI/CLIP>)

Recorded evaluation configuration: {"Backbone":"CLIP-B/32","Category":"Original"}. This record does not assert independent reproduction.

### CSR

Clean Accuracy: 86.1. Result ID: 69a17d1735dfcabb00e7612c.

[Reporting paper](<https://www.sota2.com/research/paper/69799806a2b28e761d87cb52.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69799806a2b28e761d87cb52.md>)

Recorded evaluation configuration: {"Backbone":"CLIP-B/32","Category":"Test-Time Defense"}. This record does not assert independent reproduction.

### TTE

Clean Accuracy: 86. Result ID: 69a17d1735dfcabb00e7612a.

[Reporting paper](<https://www.sota2.com/research/paper/69799806a2b28e761d87cb52.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6a711e209131cea2817e.md>)

Recorded evaluation configuration: {"Backbone":"CLIP-B/32","Category":"Test-Time Defense"}. This record does not assert independent reproduction.

### LPF

Clean Accuracy: 84.9. Result ID: 69a17d1735dfcabb00e76127.

[Reporting paper](<https://www.sota2.com/research/paper/69799806a2b28e761d87cb52.md>)

Recorded evaluation configuration: {"Backbone":"CLIP-B/32","Category":"Test-Time Defense"}. This record does not assert independent reproduction.

### Anti-Adv

Clean Accuracy: 84.4. Result ID: 69a17d1735dfcabb00e76129.

[Reporting paper](<https://www.sota2.com/research/paper/69799806a2b28e761d87cb52.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6b7f1e209131cea2e9af.md>)

Recorded evaluation configuration: {"Backbone":"CLIP-B/32","Category":"Test-Time Defense"}. This record does not assert independent reproduction.

### RanLI - Linear

Clean Accuracy: 84.16. Result ID: 6992775a3a6c3148490f278c.

[Reporting paper](<https://www.sota2.com/research/paper/68fa72bf1e209131cea585bb.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa671d1e209131cea145d5.md>)

Recorded evaluation configuration: {"Backbone":"Swin Transformer","Initialization":"Random Initialization","Finetuning Protocol":"Linear Probing"}. This record does not assert independent reproduction.

### PMG

Clean Accuracy: 82.8. Result ID: 69a17dcc35dfcabb00e7657c.

[Reporting paper](<https://www.sota2.com/research/paper/69705d96e25ad2fa9d1ed1a8.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa72df1e209131cea5912e.md>)

Recorded evaluation configuration: {"Backbone":"ViT-B/32","Proxy CLIP":"ViT-B/16","Attack Protocol":"PGD-10","Training time per epoch (s)":"1817"}. This record does not assert independent reproduction.

### R-TPT

Clean Accuracy: 76.7. Result ID: 69a17d1735dfcabb00e76126.

[Reporting paper](<https://www.sota2.com/research/paper/69799806a2b28e761d87cb52.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7add1e209131cea8590d.md>)

[Implementation](<https://github.com/TomSheng21/R-TPT>)

Recorded evaluation configuration: {"Backbone":"CLIP-B/32","Category":"Test-Time Defense"}. This record does not assert independent reproduction.

### FARE

Clean Accuracy: 76.6. Result ID: 69a17d1735dfcabb00e76282.

[Reporting paper](<https://www.sota2.com/research/paper/69799806a2b28e761d87cb52.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa738c1e209131cea5d24a.md>)

Recorded evaluation configuration: {"Backbone":"CLIP-L/14","Method category":"Adversarial Fine-tuning"}. This record does not assert independent reproduction.

### TeCoA

Clean Accuracy: 73.7. Result ID: 69a17d1735dfcabb00e76281.

[Reporting paper](<https://www.sota2.com/research/paper/69799806a2b28e761d87cb52.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6f271e209131cea44177.md>)

Recorded evaluation configuration: {"Backbone":"CLIP-L/14","Method category":"Adversarial Fine-tuning"}. This record does not assert independent reproduction.

### FT-TeCoA

Clean Accuracy: 66.79. Result ID: 69a17dcc35dfcabb00e7657b.

[Reporting paper](<https://www.sota2.com/research/paper/69705d96e25ad2fa9d1ed1a8.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6f271e209131cea44177.md>)

Recorded evaluation configuration: {"Backbone":"ViT-B/32","Proxy CLIP":"ViT-B/16","Attack Protocol":"PGD-10","Training time per epoch (s)":"1112"}. This record does not assert independent reproduction.

### TeCoA

Clean Accuracy: 50.3. Result ID: 69a17d1735dfcabb00e76124.

[Reporting paper](<https://www.sota2.com/research/paper/69799806a2b28e761d87cb52.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6f271e209131cea44177.md>)

Recorded evaluation configuration: {"Backbone":"CLIP-B/32","Category":"Adversarial Fine-tuning"}. This record does not assert independent reproduction.

### FARE

Clean Accuracy: 47.2. Result ID: 69a17d1735dfcabb00e76125.

[Reporting paper](<https://www.sota2.com/research/paper/69799806a2b28e761d87cb52.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa738c1e209131cea5d24a.md>)

Recorded evaluation configuration: {"Backbone":"CLIP-B/32","Category":"Adversarial Fine-tuning"}. This record does not assert independent reproduction.

Page 1 of 1. This is an index page, not the full collection.
