# Image Classification on Composite ImageNet Distribution Shifts (test)

Source: [Image Classification on Composite ImageNet Distribution Shifts (test)](<https://www.sota2.com/research/sota/image-classification-on-composite-imagenet-distribution-shifts-test>)

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

Data updated: 2026-02-26T05:35:13.728Z

Evaluation of Image Classification performance across a specific composite set of ImageNet distribution shift datasets (V2, R, Sketch, ObjectNet, A) using the test split.

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) | ImageNet Distribution Shifts Average of ImageNet-V2, ImageNet-R, ImageNet-Sketch, ObjectNet, and ImageNet-A (test) |
| Metric | Average Accuracy |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 19 of 19 |
| 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

- [Average Accuracy](<https://www.sota2.com/research/sota/image-classification-on-composite-imagenet-distribution-shifts-test.md?metric=Average%20Accuracy>): Higher is better; unit: PERCENT.

## Recorded results

| Method | Average Accuracy | Result date | Report paper |
| --- | --- | --- | --- |
| Greedy soup | 85.02 | Not recorded | 68fa6ce11e209131cea3679c |
| Best model in hyperparam search | 84.68 | Not recorded | 68fa6ce11e209131cea3679c |
| ViT-G (reevaluated) | 82.06 | Not recorded | 68fa6ce11e209131cea3679c |
| CLAP | 60.04 | Not recorded | 68fa72b21e209131cea57f78 |
| Zero-Shot | 57.17 | Not recorded | 68fa72b21e209131cea57f78 |
| TaskRes(e) | 55.35 | Not recorded | 68fa72b21e209131cea57f78 |
| ZS-LP | 53.65 | Not recorded | 68fa72b21e209131cea57f78 |
| CLIP-Adapter | 50.72 | Not recorded | 68fa72b21e209131cea57f78 |
| CLAP | 42.91 | Not recorded | 68fa72b21e209131cea57f78 |
| TIP-Adapter | 41.55 | Not recorded | 68fa72b21e209131cea57f78 |
| TIP-Adapter(f) | 41.36 | Not recorded | 68fa72b21e209131cea57f78 |
| TaskRes(e) | 41.28 | Not recorded | 68fa72b21e209131cea57f78 |
| TIP-Adapter | 40.69 | Not recorded | 68fa72b21e209131cea57f78 |
| Zero-Shot | 40.61 | Not recorded | 68fa72b21e209131cea57f78 |
| Rand. Init LP | 40.41 | Not recorded | 68fa72b21e209131cea57f78 |
| ZS-LP | 36.58 | Not recorded | 68fa72b21e209131cea57f78 |
| TIP-Adapter(f) | 35.58 | Not recorded | 68fa72b21e209131cea57f78 |
| CLIP-Adapter | 31.21 | Not recorded | 68fa72b21e209131cea57f78 |
| Rand. Init LP | 24.61 | Not recorded | 68fa72b21e209131cea57f78 |

### Greedy soup

Average Accuracy: 85.02. Result ID: 698f48c63a6c314849f7ab81.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6ce11e209131cea3679c.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6ce11e209131cea3679c.md>)

[Implementation](<https://github.com/mlfoundations/model-soups>)

Recorded evaluation configuration: {"Backbone":"ViT-G/14","Pre-training Dataset":"JFT-3B"}. This record does not assert independent reproduction.

### Best model in hyperparam search

Average Accuracy: 84.68. Result ID: 698f48c63a6c314849f7ab80.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6ce11e209131cea3679c.md>)

Recorded evaluation configuration: {"Backbone":"ViT-G/14","Pre-training Dataset":"JFT-3B"}. This record does not assert independent reproduction.

### ViT-G (reevaluated)

Average Accuracy: 82.06. Result ID: 698f48c63a6c314849f7ab7f.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6ce11e209131cea3679c.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6cd91e209131cea363c2.md>)

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

Recorded evaluation configuration: {"Backbone":"ViT-G/14","Pre-training Dataset":"JFT-3B"}. This record does not assert independent reproduction.

### CLAP

Average Accuracy: 60.04. Result ID: 699251b73a6c3148490e26da.

[Reporting paper](<https://www.sota2.com/research/paper/68fa72b21e209131cea57f78.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa72b21e209131cea57f78.md>)

[Implementation](<https://github.com/jusiro/CLAP>)

Recorded evaluation configuration: {"Backbone":"ViT-B/16"}. This record does not assert independent reproduction.

### Zero-Shot

Average Accuracy: 57.17. Result ID: 699251b73a6c3148490e26d3.

[Reporting paper](<https://www.sota2.com/research/paper/68fa72b21e209131cea57f78.md>)

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

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

Recorded evaluation configuration: {"Backbone":"ViT-B/16"}. This record does not assert independent reproduction.

### TaskRes(e)

Average Accuracy: 55.35. Result ID: 699251b73a6c3148490e26d8.

[Reporting paper](<https://www.sota2.com/research/paper/68fa72b21e209131cea57f78.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6ee41e209131cea42ac8.md>)

Recorded evaluation configuration: {"Backbone":"ViT-B/16"}. This record does not assert independent reproduction.

### ZS-LP

Average Accuracy: 53.65. Result ID: 699251b73a6c3148490e26d9.

[Reporting paper](<https://www.sota2.com/research/paper/68fa72b21e209131cea57f78.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa72b21e209131cea57f78.md>)

[Implementation](<https://github.com/jusiro/CLAP>)

Recorded evaluation configuration: {"Backbone":"ViT-B/16"}. This record does not assert independent reproduction.

### CLIP-Adapter

Average Accuracy: 50.72. Result ID: 699251b73a6c3148490e26d5.

[Reporting paper](<https://www.sota2.com/research/paper/68fa72b21e209131cea57f78.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa77921e209131cea7403b.md>)

Recorded evaluation configuration: {"Backbone":"ViT-B/16"}. This record does not assert independent reproduction.

### CLAP

Average Accuracy: 42.91. Result ID: 699251b73a6c3148490e26d2.

[Reporting paper](<https://www.sota2.com/research/paper/68fa72b21e209131cea57f78.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa72b21e209131cea57f78.md>)

[Implementation](<https://github.com/jusiro/CLAP>)

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

### TIP-Adapter

Average Accuracy: 41.55. Result ID: 699251b73a6c3148490e26d6.

[Reporting paper](<https://www.sota2.com/research/paper/68fa72b21e209131cea57f78.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6b441e209131cea2d1aa.md>)

Recorded evaluation configuration: {"Backbone":"ViT-B/16"}. This record does not assert independent reproduction.

### TIP-Adapter(f)

Average Accuracy: 41.36. Result ID: 699251b73a6c3148490e26cf.

[Reporting paper](<https://www.sota2.com/research/paper/68fa72b21e209131cea57f78.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6b441e209131cea2d1aa.md>)

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

### TaskRes(e)

Average Accuracy: 41.28. Result ID: 699251b73a6c3148490e26d0.

[Reporting paper](<https://www.sota2.com/research/paper/68fa72b21e209131cea57f78.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6ee41e209131cea42ac8.md>)

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

### TIP-Adapter

Average Accuracy: 40.69. Result ID: 699251b73a6c3148490e26ce.

[Reporting paper](<https://www.sota2.com/research/paper/68fa72b21e209131cea57f78.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6b441e209131cea2d1aa.md>)

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

### Zero-Shot

Average Accuracy: 40.61. Result ID: 699251b73a6c3148490e26cb.

[Reporting paper](<https://www.sota2.com/research/paper/68fa72b21e209131cea57f78.md>)

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

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

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

### Rand. Init LP

Average Accuracy: 40.41. Result ID: 699251b73a6c3148490e26d4.

[Reporting paper](<https://www.sota2.com/research/paper/68fa72b21e209131cea57f78.md>)

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

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

Recorded evaluation configuration: {"Backbone":"ViT-B/16"}. This record does not assert independent reproduction.

### ZS-LP

Average Accuracy: 36.58. Result ID: 699251b73a6c3148490e26d1.

[Reporting paper](<https://www.sota2.com/research/paper/68fa72b21e209131cea57f78.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa72b21e209131cea57f78.md>)

[Implementation](<https://github.com/jusiro/CLAP>)

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

### TIP-Adapter(f)

Average Accuracy: 35.58. Result ID: 699251b73a6c3148490e26d7.

[Reporting paper](<https://www.sota2.com/research/paper/68fa72b21e209131cea57f78.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6b441e209131cea2d1aa.md>)

Recorded evaluation configuration: {"Backbone":"ViT-B/16"}. This record does not assert independent reproduction.

### CLIP-Adapter

Average Accuracy: 31.21. Result ID: 699251b73a6c3148490e26cd.

[Reporting paper](<https://www.sota2.com/research/paper/68fa72b21e209131cea57f78.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa77921e209131cea7403b.md>)

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

### Rand. Init LP

Average Accuracy: 24.61. Result ID: 699251b73a6c3148490e26cc.

[Reporting paper](<https://www.sota2.com/research/paper/68fa72b21e209131cea57f78.md>)

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

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

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

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