# Visual Adaptation on VTAB 1k (test)

Source: [Visual Adaptation on VTAB 1k (test)](<https://www.sota2.com/research/sota/visual-adaptation-on-vtab-1k-test>)

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

Data updated: 2026-02-26T05:53:33.323Z

Evaluation of model performance on the test split of the VTAB 1k benchmark, reporting mean accuracy across the Natural, Specialized, and Structured task groupings.

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 | Visual Adaptation |
| Dataset (including any stated split) | VTAB 1k (test) |
| Metric | Natural Accuracy |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 16 of 16 |
| 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

- [Natural Accuracy](<https://www.sota2.com/research/sota/visual-adaptation-on-vtab-1k-test.md?metric=Natural%20Accuracy>): Higher is better; unit: PERCENT.

- [Specialized Accuracy](<https://www.sota2.com/research/sota/visual-adaptation-on-vtab-1k-test.md?metric=Specialized%20Accuracy>): Higher is better; unit: PERCENT.

- [Structured Accuracy](<https://www.sota2.com/research/sota/visual-adaptation-on-vtab-1k-test.md?metric=Structured%20Accuracy>): Higher is better; unit: PERCENT.

- [Mean Total Accuracy](<https://www.sota2.com/research/sota/visual-adaptation-on-vtab-1k-test.md?metric=Mean%20Total%20Accuracy>): Higher is better; unit: PERCENT.

## Recorded results

| Method | Natural Accuracy | Result date | Report paper |
| --- | --- | --- | --- |
| VPT-Deep | 82.5 | Not recorded | 68fa71dd1e209131cea536f1 |
| ARC | 82.3 | Not recorded | 68fa71dd1e209131cea536f1 |
| LORA | 81.4 | Not recorded | 68fa71dd1e209131cea536f1 |
| ARC | 79.1 | Not recorded | 68fa71dd1e209131cea536f1 |
| VPT-Shallow | 78.7 | Not recorded | 68fa71dd1e209131cea536f1 |
| VPT-Deep | 77.9 | Not recorded | 68fa71dd1e209131cea536f1 |
| LORA | 77.1 | Not recorded | 68fa71dd1e209131cea536f1 |
| VPT-Shallow | 74.8 | Not recorded | 68fa71dd1e209131cea536f1 |
| Full fine-tuning | 74.7 | Not recorded | 68fa71dd1e209131cea536f1 |
| Linear probing | 70.9 | Not recorded | 68fa71dd1e209131cea536f1 |
| Full fine-tuning | 70.9 | Not recorded | 68fa71dd1e209131cea536f1 |
| Bias | 70.5 | Not recorded | 68fa71dd1e209131cea536f1 |
| Bias | 70.3 | Not recorded | 68fa71dd1e209131cea536f1 |
| Adapter | 68.6 | Not recorded | 68fa71dd1e209131cea536f1 |
| Adapter | 68.1 | Not recorded | 68fa71dd1e209131cea536f1 |
| Linear probing | 67.9 | Not recorded | 68fa71dd1e209131cea536f1 |

### VPT-Deep

Natural Accuracy: 82.5. Result ID: 6991e9ad3a6c3148490b670b.

[Reporting paper](<https://www.sota2.com/research/paper/68fa71dd1e209131cea536f1.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":"ViT-Large","Pre-trained":"ImageNet-21k","Params.":"0.49"}. This record does not assert independent reproduction.

### ARC

Natural Accuracy: 82.3. Result ID: 6991e9ad3a6c3148490b670d.

[Reporting paper](<https://www.sota2.com/research/paper/68fa71dd1e209131cea536f1.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa71dd1e209131cea536f1.md>)

[Implementation](<https://github.com/DavidYanAnDe/ARC>)

Recorded evaluation configuration: {"Backbone":"ViT-Large","Pre-trained":"ImageNet-21k","Params.":"0.18"}. This record does not assert independent reproduction.

### LORA

Natural Accuracy: 81.4. Result ID: 6991e9ad3a6c3148490b670c.

[Reporting paper](<https://www.sota2.com/research/paper/68fa71dd1e209131cea536f1.md>)

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

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

Recorded evaluation configuration: {"Backbone":"ViT-Large","Pre-trained":"ImageNet-21k","Params.":"0.74"}. This record does not assert independent reproduction.

### ARC

Natural Accuracy: 79.1. Result ID: 6991e9ad3a6c3148490b671e.

[Reporting paper](<https://www.sota2.com/research/paper/68fa71dd1e209131cea536f1.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa71dd1e209131cea536f1.md>)

[Implementation](<https://github.com/DavidYanAnDe/ARC>)

Recorded evaluation configuration: {"Backbone":"ViT-Huge","Pre-trained":"ImageNet-21k","Params.":"0.22"}. This record does not assert independent reproduction.

### VPT-Shallow

Natural Accuracy: 78.7. Result ID: 6991e9ad3a6c3148490b670a.

[Reporting paper](<https://www.sota2.com/research/paper/68fa71dd1e209131cea536f1.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":"ViT-Large","Pre-trained":"ImageNet-21k","Params.":"0.15"}. This record does not assert independent reproduction.

### VPT-Deep

Natural Accuracy: 77.9. Result ID: 6991e9ad3a6c3148490b671c.

[Reporting paper](<https://www.sota2.com/research/paper/68fa71dd1e209131cea536f1.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":"ViT-Huge","Pre-trained":"ImageNet-21k","Params.":"0.96"}. This record does not assert independent reproduction.

### LORA

Natural Accuracy: 77.1. Result ID: 6991e9ad3a6c3148490b671d.

[Reporting paper](<https://www.sota2.com/research/paper/68fa71dd1e209131cea536f1.md>)

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

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

Recorded evaluation configuration: {"Backbone":"ViT-Huge","Pre-trained":"ImageNet-21k","Params.":"1.21"}. This record does not assert independent reproduction.

### VPT-Shallow

Natural Accuracy: 74.8. Result ID: 6991e9ad3a6c3148490b671b.

[Reporting paper](<https://www.sota2.com/research/paper/68fa71dd1e209131cea536f1.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":"ViT-Huge","Pre-trained":"ImageNet-21k","Params.":"0.18"}. This record does not assert independent reproduction.

### Full fine-tuning

Natural Accuracy: 74.7. Result ID: 6991e9ad3a6c3148490b6706.

[Reporting paper](<https://www.sota2.com/research/paper/68fa71dd1e209131cea536f1.md>)

Recorded evaluation configuration: {"Backbone":"ViT-Large","Pre-trained":"ImageNet-21k","Params.":"303.4"}. This record does not assert independent reproduction.

### Linear probing

Natural Accuracy: 70.9. Result ID: 6991e9ad3a6c3148490b6707.

[Reporting paper](<https://www.sota2.com/research/paper/68fa71dd1e209131cea536f1.md>)

Recorded evaluation configuration: {"Backbone":"ViT-Large","Pre-trained":"ImageNet-21k","Params.":"0.05"}. This record does not assert independent reproduction.

### Full fine-tuning

Natural Accuracy: 70.9. Result ID: 6991e9ad3a6c3148490b6717.

[Reporting paper](<https://www.sota2.com/research/paper/68fa71dd1e209131cea536f1.md>)

Recorded evaluation configuration: {"Backbone":"ViT-Huge","Pre-trained":"ImageNet-21k","Params.":"630.9"}. This record does not assert independent reproduction.

### Bias

Natural Accuracy: 70.5. Result ID: 6991e9ad3a6c3148490b6709.

[Reporting paper](<https://www.sota2.com/research/paper/68fa71dd1e209131cea536f1.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6d5a1e209131cea39632.md>)

[Implementation](<https://www.github.com/benzakenelad/BitFit>)

Recorded evaluation configuration: {"Backbone":"ViT-Large","Pre-trained":"ImageNet-21k","Params.":"0.32"}. This record does not assert independent reproduction.

### Bias

Natural Accuracy: 70.3. Result ID: 6991e9ad3a6c3148490b671a.

[Reporting paper](<https://www.sota2.com/research/paper/68fa71dd1e209131cea536f1.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6d5a1e209131cea39632.md>)

[Implementation](<https://www.github.com/benzakenelad/BitFit>)

Recorded evaluation configuration: {"Backbone":"ViT-Huge","Pre-trained":"ImageNet-21k","Params.":"0.52"}. This record does not assert independent reproduction.

### Adapter

Natural Accuracy: 68.6. Result ID: 6991e9ad3a6c3148490b6708.

[Reporting paper](<https://www.sota2.com/research/paper/68fa71dd1e209131cea536f1.md>)

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

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

Recorded evaluation configuration: {"Backbone":"ViT-Large","Pre-trained":"ImageNet-21k","Params.":"2.38"}. This record does not assert independent reproduction.

### Adapter

Natural Accuracy: 68.1. Result ID: 6991e9ad3a6c3148490b6719.

[Reporting paper](<https://www.sota2.com/research/paper/68fa71dd1e209131cea536f1.md>)

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

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

Recorded evaluation configuration: {"Backbone":"ViT-Huge","Pre-trained":"ImageNet-21k","Params.":"5.78"}. This record does not assert independent reproduction.

### Linear probing

Natural Accuracy: 67.9. Result ID: 6991e9ad3a6c3148490b6718.

[Reporting paper](<https://www.sota2.com/research/paper/68fa71dd1e209131cea536f1.md>)

Recorded evaluation configuration: {"Backbone":"ViT-Huge","Pre-trained":"ImageNet-21k","Params.":"0.06"}. This record does not assert independent reproduction.

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