# Image Classification on ImageNet-1K (ImageNet-21K Pre-training) (test)

Source: [Image Classification on ImageNet-1K (ImageNet-21K Pre-training) (test)](<https://www.sota2.com/research/sota/image-classification-on-imagenet-1k-imagenet-21k-pre-training-test>)

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

Data updated: 2026-02-26T06:06:38.108Z

Evaluation of Image Classification performance on the ImageNet-1K test set, specifically for models that utilized ImageNet-21K for pre-training.

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-1K fine-tuning after ImageNet-21K pre-training (test) |
| Metric | Top-1 Acc |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 12 of 12 |
| 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

- [Top-1 Acc](<https://www.sota2.com/research/sota/image-classification-on-imagenet-1k-imagenet-21k-pre-training-test.md?metric=Top-1%20Acc>): Higher is better; unit: PERCENT.

## Recorded results

| Method | Top-1 Acc | Result date | Report paper |
| --- | --- | --- | --- |
| CSWin-L | 87.5 | Not recorded | 68fa6b991e209131cea2f30f |
| Swin-L | 87.3 | Not recorded | 68fa6b991e209131cea2f30f |
| CSWin-B | 87 | Not recorded | 68fa6b991e209131cea2f30f |
| CSWin-L | 86.5 | Not recorded | 68fa6b991e209131cea2f30f |
| Swin-B | 86.4 | Not recorded | 68fa6b991e209131cea2f30f |
| Swin-L | 86.3 | Not recorded | 68fa6b991e209131cea2f30f |
| CSWin-B | 85.9 | Not recorded | 68fa6b991e209131cea2f30f |
| R-152x4 | 85.4 | Not recorded | 68fa6b991e209131cea2f30f |
| ViT-L/16 | 85.2 | Not recorded | 68fa6b991e209131cea2f30f |
| Swin-B | 85.2 | Not recorded | 68fa6b991e209131cea2f30f |
| R-101x3 | 84.4 | Not recorded | 68fa6b991e209131cea2f30f |
| ViT-B/16 | 84 | Not recorded | 68fa6b991e209131cea2f30f |

### CSWin-L

Top-1 Acc: 87.5. Result ID: 698e98a5ffc3901de09efb3b.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6b991e209131cea2f30f.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6b991e209131cea2f30f.md>)

Recorded evaluation configuration: {"Param":"173M","Size":"384x384","FLOPs":"96.8G","Pre-training":"ImageNet-21K"}. This record does not assert independent reproduction.

### Swin-L

Top-1 Acc: 87.3. Result ID: 698e98a5ffc3901de09efb37.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6b991e209131cea2f30f.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6a8c1e209131cea28b62.md>)

[Implementation](<https://github.com/microsoft/Swin-Transformer>)

Recorded evaluation configuration: {"Param":"197M","Size":"384x384","FLOPs":"103.9G","Pre-training":"ImageNet-21K"}. This record does not assert independent reproduction.

### CSWin-B

Top-1 Acc: 87. Result ID: 698e98a5ffc3901de09efb39.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6b991e209131cea2f30f.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6b991e209131cea2f30f.md>)

Recorded evaluation configuration: {"Param":"78M","Size":"384x384","FLOPs":"47.0G","Pre-training":"ImageNet-21K"}. This record does not assert independent reproduction.

### CSWin-L

Top-1 Acc: 86.5. Result ID: 698e98a5ffc3901de09efb3a.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6b991e209131cea2f30f.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6b991e209131cea2f30f.md>)

Recorded evaluation configuration: {"Param":"173M","Size":"224x224","FLOPs":"31.5G","Pre-training":"ImageNet-21K"}. This record does not assert independent reproduction.

### Swin-B

Top-1 Acc: 86.4. Result ID: 698e98a5ffc3901de09efb35.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6b991e209131cea2f30f.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6a8c1e209131cea28b62.md>)

[Implementation](<https://github.com/microsoft/Swin-Transformer>)

Recorded evaluation configuration: {"Param":"88M","Size":"384x384","FLOPs":"47.1G","Pre-training":"ImageNet-21K"}. This record does not assert independent reproduction.

### Swin-L

Top-1 Acc: 86.3. Result ID: 698e98a5ffc3901de09efb36.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6b991e209131cea2f30f.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6a8c1e209131cea28b62.md>)

[Implementation](<https://github.com/microsoft/Swin-Transformer>)

Recorded evaluation configuration: {"Param":"197M","Size":"224x224","FLOPs":"34.5G","Pre-training":"ImageNet-21K"}. This record does not assert independent reproduction.

### CSWin-B

Top-1 Acc: 85.9. Result ID: 698e98a5ffc3901de09efb38.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6b991e209131cea2f30f.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6b991e209131cea2f30f.md>)

Recorded evaluation configuration: {"Param":"78M","Size":"224x224","FLOPs":"15.0G","Pre-training":"ImageNet-21K"}. This record does not assert independent reproduction.

### R-152x4

Top-1 Acc: 85.4. Result ID: 698e98a5ffc3901de09efb31.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6b991e209131cea2f30f.md>)

Recorded evaluation configuration: {"Param":"937M","Size":"480x480","FLOPs":"840.5G","Pre-training":"ImageNet-21K"}. This record does not assert independent reproduction.

### ViT-L/16

Top-1 Acc: 85.2. Result ID: 698e98a5ffc3901de09efb33.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6b991e209131cea2f30f.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6a0b1e209131cea25b04.md>)

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

Recorded evaluation configuration: {"Param":"307M","Size":"384x384","FLOPs":"190.7G","Pre-training":"ImageNet-21K"}. This record does not assert independent reproduction.

### Swin-B

Top-1 Acc: 85.2. Result ID: 698e98a5ffc3901de09efb34.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6b991e209131cea2f30f.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6a8c1e209131cea28b62.md>)

[Implementation](<https://github.com/microsoft/Swin-Transformer>)

Recorded evaluation configuration: {"Param":"88M","Size":"224x224","FLOPs":"15.4G","Pre-training":"ImageNet-21K"}. This record does not assert independent reproduction.

### R-101x3

Top-1 Acc: 84.4. Result ID: 698e98a5ffc3901de09efb30.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6b991e209131cea2f30f.md>)

Recorded evaluation configuration: {"Param":"388M","Size":"384x384","FLOPs":"204.6G","Pre-training":"ImageNet-21K"}. This record does not assert independent reproduction.

### ViT-B/16

Top-1 Acc: 84. Result ID: 698e98a5ffc3901de09efb32.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6b991e209131cea2f30f.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6a0b1e209131cea25b04.md>)

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

Recorded evaluation configuration: {"Param":"86M","Size":"384x384","FLOPs":"55.4G","Pre-training":"ImageNet-21K"}. This record does not assert independent reproduction.

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