# Image Classification on ImageNet (val) (Accuracy and Throughput)

Source: [Image Classification on ImageNet (val) (Accuracy and Throughput)](<https://www.sota2.com/research/sota/image-classification-on-imagenet-val-accuracy-and-throughput>)

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

Data updated: 2026-02-26T05:33:57.497Z

Evaluation of Image Classification performance on the validation split of the ImageNet dataset, reporting standard accuracy and inference throughput (images per second).

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 (val) |
| Metric | Accuracy (%) |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 27 of 27 |
| 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

- [Accuracy (%)](<https://www.sota2.com/research/sota/image-classification-on-imagenet-val-accuracy-and-throughput.md?metric=Accuracy%20(%25)>): Higher is better; unit: PERCENT.

- [Throughput (img/sec)](<https://www.sota2.com/research/sota/image-classification-on-imagenet-val-accuracy-and-throughput.md?metric=Throughput%20(img%2Fsec)>): Higher is better; unit: SCALAR.

## Recorded results

| Method | Accuracy (%) | Result date | Report paper |
| --- | --- | --- | --- |
| ResNeSt-269 | 84.5 | Not recorded | 68fa68f91e209131cea1f68a |
| EfficientNet-B7 | 84.4 | Not recorded | 68fa68f91e209131cea1f68a |
| GPipe | 84.3 | Not recorded | 68fa68f91e209131cea1f68a |
| SigLIP 2 | 84.1 | Not recorded | 698c0d29c97a0b3e3b5e1223 |
| ResNeSt-200 | 83.9 | Not recorded | 68fa68f91e209131cea1f68a |
| EfficientNet-B5 | 83.7 | Not recorded | 68fa68f91e209131cea1f68a |
| AmoebaNet-C | 83.5 | Not recorded | 68fa68f91e209131cea1f68a |
| EVA-CLIP-8B | 83.5 | Not recorded | 698c0d29c97a0b3e3b5e1223 |
| SigLIP | 83.2 | Not recorded | 698c0d29c97a0b3e3b5e1223 |
| ResNeSt-101 | 83 | Not recorded | 68fa68f91e209131cea1f68a |
| EfficientNet-B4 | 83 | Not recorded | 68fa68f91e209131cea1f68a |
| AmoebaNet-A | 82.8 | Not recorded | 68fa68f91e209131cea1f68a |
| SENet-154 | 82.7 | Not recorded | 68fa68f91e209131cea1f68a |
| NASNet-A | 82.7 | Not recorded | 68fa68f91e209131cea1f68a |
| DINOv3 | 82.3 | Not recorded | 698c0d29c97a0b3e3b5e1223 |
| VersaViT\* | 75.6 | Not recorded | 698c0d29c97a0b3e3b5e1223 |
| CLIP-L | 75.5 | Not recorded | 698c0d29c97a0b3e3b5e1223 |
| CoOp | 71.51 | Not recorded | 6937acfe6cfd488761a4e701 |
| RMAdapter | 71.37 | Not recorded | 6937acfe6cfd488761a4e701 |
| PromptSRC | 71.27 | Not recorded | 6937acfe6cfd488761a4e701 |
| KgCoOp | 71.2 | Not recorded | 6937acfe6cfd488761a4e701 |
| Co-CoOp | 71.02 | Not recorded | 6937acfe6cfd488761a4e701 |
| MMA | 71 | Not recorded | 6937acfe6cfd488761a4e701 |
| CoPrompt | 70.8 | Not recorded | 6937acfe6cfd488761a4e701 |
| MaPLe | 70.72 | Not recorded | 6937acfe6cfd488761a4e701 |
| VersaViT | 67.9 | Not recorded | 698c0d29c97a0b3e3b5e1223 |
| CLIP | 66.73 | Not recorded | 6937acfe6cfd488761a4e701 |

### ResNeSt-269

Accuracy (%): 84.5. Result ID: 698d91e08b517eb4332267c3.

[Reporting paper](<https://www.sota2.com/research/paper/68fa68f91e209131cea1f68a.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa68f91e209131cea1f68a.md>)

Recorded evaluation configuration: {"Parameters":"111M","Crop Size":416,"Hardware":"NVIDIA V100 GPU","mini-batch size":16}. This record does not assert independent reproduction.

### EfficientNet-B7

Accuracy (%): 84.4. Result ID: 698d91e08b517eb4332267c5.

[Reporting paper](<https://www.sota2.com/research/paper/68fa68f91e209131cea1f68a.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa684f1e209131cea1b521.md>)

[Implementation](<https://github.com/tensorflow/tpu/tree/master/models/official/efficientnet>)

Recorded evaluation configuration: {"Parameters":"66M","Crop Size":600,"Hardware":"NVIDIA V100 GPU","mini-batch size":16}. This record does not assert independent reproduction.

### GPipe

Accuracy (%): 84.3. Result ID: 698d91e08b517eb4332267c4.

[Reporting paper](<https://www.sota2.com/research/paper/68fa68f91e209131cea1f68a.md>)

Recorded evaluation configuration: {"Parameters":"557M","Hardware":"NVIDIA V100 GPU","mini-batch size":16}. This record does not assert independent reproduction.

### SigLIP 2

Accuracy (%): 84.1. Result ID: 69971a875698665ec4687005.

[Reporting paper](<https://www.sota2.com/research/paper/698c0d29c97a0b3e3b5e1223.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa770a1e209131cea70ad4.md>)

Recorded evaluation configuration: {"Zero-shot":true}. This record does not assert independent reproduction.

### ResNeSt-200

Accuracy (%): 83.9. Result ID: 698d91e08b517eb4332267c0.

[Reporting paper](<https://www.sota2.com/research/paper/68fa68f91e209131cea1f68a.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa68f91e209131cea1f68a.md>)

Recorded evaluation configuration: {"Parameters":"70M","Crop Size":320,"Hardware":"NVIDIA V100 GPU","mini-batch size":16}. This record does not assert independent reproduction.

### EfficientNet-B5

Accuracy (%): 83.7. Result ID: 698d91e08b517eb4332267c1.

[Reporting paper](<https://www.sota2.com/research/paper/68fa68f91e209131cea1f68a.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa684f1e209131cea1b521.md>)

[Implementation](<https://github.com/tensorflow/tpu/tree/master/models/official/efficientnet>)

Recorded evaluation configuration: {"Parameters":"30M","Crop Size":456,"Hardware":"NVIDIA V100 GPU","mini-batch size":16}. This record does not assert independent reproduction.

### AmoebaNet-C

Accuracy (%): 83.5. Result ID: 698d91e08b517eb4332267c2.

[Reporting paper](<https://www.sota2.com/research/paper/68fa68f91e209131cea1f68a.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa65691e209131cea0a262.md>)

[Implementation](<https://colab.research.google.com/github/google-research/google-research/blob/master/evolution/regularized_evolution_algorithm/regularized_evolution.ipynb>)

Recorded evaluation configuration: {"Parameters":"155M","Crop Size":299,"Hardware":"NVIDIA V100 GPU","mini-batch size":16}. This record does not assert independent reproduction.

### EVA-CLIP-8B

Accuracy (%): 83.5. Result ID: 69971a875698665ec4687003.

[Reporting paper](<https://www.sota2.com/research/paper/698c0d29c97a0b3e3b5e1223.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa72141e209131cea545f5.md>)

[Implementation](<https://www.sota2.com/baaivision/EVA/EVA-CLIP-18B>)

Recorded evaluation configuration: {"Zero-shot":true}. This record does not assert independent reproduction.

### SigLIP

Accuracy (%): 83.2. Result ID: 69971a875698665ec4687004.

[Reporting paper](<https://www.sota2.com/research/paper/698c0d29c97a0b3e3b5e1223.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa70ae1e209131cea4c7a4.md>)

Recorded evaluation configuration: {"Zero-shot":true}. This record does not assert independent reproduction.

### ResNeSt-101

Accuracy (%): 83. Result ID: 698d91e08b517eb4332267bb.

[Reporting paper](<https://www.sota2.com/research/paper/68fa68f91e209131cea1f68a.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa68f91e209131cea1f68a.md>)

Recorded evaluation configuration: {"Parameters":"48M","Crop Size":256,"Hardware":"NVIDIA V100 GPU","mini-batch size":16}. This record does not assert independent reproduction.

### EfficientNet-B4

Accuracy (%): 83. Result ID: 698d91e08b517eb4332267bc.

[Reporting paper](<https://www.sota2.com/research/paper/68fa68f91e209131cea1f68a.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa684f1e209131cea1b521.md>)

[Implementation](<https://github.com/tensorflow/tpu/tree/master/models/official/efficientnet>)

Recorded evaluation configuration: {"Parameters":"19M","Crop Size":380,"Hardware":"NVIDIA V100 GPU","mini-batch size":16}. This record does not assert independent reproduction.

### AmoebaNet-A

Accuracy (%): 82.8. Result ID: 698d91e08b517eb4332267bf.

[Reporting paper](<https://www.sota2.com/research/paper/68fa68f91e209131cea1f68a.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa65691e209131cea0a262.md>)

[Implementation](<https://colab.research.google.com/github/google-research/google-research/blob/master/evolution/regularized_evolution_algorithm/regularized_evolution.ipynb>)

Recorded evaluation configuration: {"Parameters":"87M","Crop Size":299,"Hardware":"NVIDIA V100 GPU","mini-batch size":16}. This record does not assert independent reproduction.

### SENet-154

Accuracy (%): 82.7. Result ID: 698d91e08b517eb4332267bd.

[Reporting paper](<https://www.sota2.com/research/paper/68fa68f91e209131cea1f68a.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa65bd1e209131cea0c0f9.md>)

[Implementation](<https://github.com/hujie-frank/SENet>)

Recorded evaluation configuration: {"Parameters":"146M","Crop Size":320,"Hardware":"NVIDIA V100 GPU","mini-batch size":16}. This record does not assert independent reproduction.

### NASNet-A

Accuracy (%): 82.7. Result ID: 698d91e08b517eb4332267be.

[Reporting paper](<https://www.sota2.com/research/paper/68fa68f91e209131cea1f68a.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa64441e209131cea039e2.md>)

Recorded evaluation configuration: {"Parameters":"89M","Crop Size":331,"Hardware":"NVIDIA V100 GPU","mini-batch size":16}. This record does not assert independent reproduction.

### DINOv3

Accuracy (%): 82.3. Result ID: 69971a875698665ec4687006.

[Reporting paper](<https://www.sota2.com/research/paper/698c0d29c97a0b3e3b5e1223.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7a931e209131cea83fc5.md>)

Recorded evaluation configuration: {"Zero-shot":true}. This record does not assert independent reproduction.

### VersaViT\*

Accuracy (%): 75.6. Result ID: 69971a875698665ec4687008.

[Reporting paper](<https://www.sota2.com/research/paper/698c0d29c97a0b3e3b5e1223.md>)

[Method origin paper](<https://www.sota2.com/research/paper/698c0d29c97a0b3e3b5e1223.md>)

Recorded evaluation configuration: {"Zero-shot":true,"Alignment":"30M image-text pairs"}. This record does not assert independent reproduction.

### CLIP-L

Accuracy (%): 75.5. Result ID: 69971a875698665ec4687002.

[Reporting paper](<https://www.sota2.com/research/paper/698c0d29c97a0b3e3b5e1223.md>)

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

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

Recorded evaluation configuration: {"Zero-shot":true}. This record does not assert independent reproduction.

### CoOp

Accuracy (%): 71.51. Result ID: 69987e395698665ec4df3c6d.

[Reporting paper](<https://www.sota2.com/research/paper/6937acfe6cfd488761a4e701.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6d8b1e209131cea3a814.md>)

[Implementation](<https://github.com/KaiyangZhou/CoOp>)

Recorded evaluation configuration: {"mode":"Few-shot adaptation"}. This record does not assert independent reproduction.

### RMAdapter

Accuracy (%): 71.37. Result ID: 69987e395698665ec4df3c74.

[Reporting paper](<https://www.sota2.com/research/paper/6937acfe6cfd488761a4e701.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6937acfe6cfd488761a4e701.md>)

Recorded evaluation configuration: {"mode":"Few-shot adaptation"}. This record does not assert independent reproduction.

### PromptSRC

Accuracy (%): 71.27. Result ID: 69987e395698665ec4df3c71.

[Reporting paper](<https://www.sota2.com/research/paper/6937acfe6cfd488761a4e701.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa705e1e209131cea4aa58.md>)

Recorded evaluation configuration: {"mode":"Few-shot adaptation"}. This record does not assert independent reproduction.

### KgCoOp

Accuracy (%): 71.2. Result ID: 69987e395698665ec4df3c6f.

[Reporting paper](<https://www.sota2.com/research/paper/6937acfe6cfd488761a4e701.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6ee91e209131cea42d7a.md>)

Recorded evaluation configuration: {"mode":"Few-shot adaptation"}. This record does not assert independent reproduction.

### Co-CoOp

Accuracy (%): 71.02. Result ID: 69987e395698665ec4df3c6e.

[Reporting paper](<https://www.sota2.com/research/paper/6937acfe6cfd488761a4e701.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6d8c1e209131cea3a831.md>)

[Implementation](<https://github.com/KaiyangZhou/CoOp>)

Recorded evaluation configuration: {"mode":"Few-shot adaptation"}. This record does not assert independent reproduction.

### MMA

Accuracy (%): 71. Result ID: 69987e395698665ec4df3c72.

[Reporting paper](<https://www.sota2.com/research/paper/6937acfe6cfd488761a4e701.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa74c21e209131cea6392b.md>)

Recorded evaluation configuration: {"mode":"Few-shot adaptation"}. This record does not assert independent reproduction.

### CoPrompt

Accuracy (%): 70.8. Result ID: 69987e395698665ec4df3c73.

[Reporting paper](<https://www.sota2.com/research/paper/6937acfe6cfd488761a4e701.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa74641e209131cea616f7.md>)

[Implementation](<https://github.com/ShuvenduRoy/CoPrompt>)

Recorded evaluation configuration: {"mode":"Few-shot adaptation"}. This record does not assert independent reproduction.

### MaPLe

Accuracy (%): 70.72. Result ID: 69987e395698665ec4df3c70.

[Reporting paper](<https://www.sota2.com/research/paper/6937acfe6cfd488761a4e701.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6ef21e209131cea4311c.md>)

[Implementation](<https://github.com/muzairkhattak/multimodal-prompt-learning>)

Recorded evaluation configuration: {"mode":"Few-shot adaptation"}. This record does not assert independent reproduction.

### VersaViT

Accuracy (%): 67.9. Result ID: 69971a875698665ec4687007.

[Reporting paper](<https://www.sota2.com/research/paper/698c0d29c97a0b3e3b5e1223.md>)

[Method origin paper](<https://www.sota2.com/research/paper/698c0d29c97a0b3e3b5e1223.md>)

Recorded evaluation configuration: {"Zero-shot":true,"Alignment":"30M image-text pairs"}. This record does not assert independent reproduction.

### CLIP

Accuracy (%): 66.73. Result ID: 69987e395698665ec4df3c6c.

[Reporting paper](<https://www.sota2.com/research/paper/6937acfe6cfd488761a4e701.md>)

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

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

Recorded evaluation configuration: {"mode":"Zero-shot"}. This record does not assert independent reproduction.

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