# Image Classification on ImageNet (val) (Search and Retrain Accuracy)

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

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

Data updated: 2026-02-26T06:02:10.281Z

Evaluation of Image Classification performance on the validation split of the ImageNet dataset, specifically reporting accuracy derived from both search and retraining protocols.

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 | Retrain 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

- [Retrain Top-1 Acc](<https://www.sota2.com/research/sota/image-classification-on-imagenet-val-search-and-retrain-accuracy.md?metric=Retrain%20Top-1%20Acc>): Higher is better; unit: PERCENT.

- [Search Top-1 Acc](<https://www.sota2.com/research/sota/image-classification-on-imagenet-val-search-and-retrain-accuracy.md?metric=Search%20Top-1%20Acc>): Higher is better; unit: PERCENT.

- [Search Top-5 Acc](<https://www.sota2.com/research/sota/image-classification-on-imagenet-val-search-and-retrain-accuracy.md?metric=Search%20Top-5%20Acc>): Higher is better; unit: PERCENT.

- [Retrain Top-5 Acc](<https://www.sota2.com/research/sota/image-classification-on-imagenet-val-search-and-retrain-accuracy.md?metric=Retrain%20Top-5%20Acc>): Higher is better; unit: PERCENT.

## Recorded results

| Method | Retrain Top-1 Acc | Result date | Report paper |
| --- | --- | --- | --- |
| MnasNet-A1 | 75.2 | Not recorded | 68fa67551e209131cea158f4 |
| Proxyless-R (mobile) | 74.6 | Not recorded | 68fa67551e209131cea158f4 |
| Single Path One-Shot | 74.3 | Not recorded | 68fa67551e209131cea158f4 |
| PNASNET | 74.2 | Not recorded | 68fa67551e209131cea158f4 |
| NASNET-A(4@1056) | 74 | Not recorded | 68fa67551e209131cea158f4 |
| DARTS | 73.3 | Not recorded | 68fa67551e209131cea158f4 |
| SNAS | 72.7 | Not recorded | 68fa67551e209131cea158f4 |
| ShuffleNet V2 (1.5x) | 72.6 | Not recorded | 68fa67551e209131cea158f4 |
| MobileNet V2 (1.0x) | 72 | Not recorded | 68fa67551e209131cea158f4 |
| MobileNet V1 (0.75x) | 68.4 | Not recorded | 68fa67551e209131cea158f4 |

### MnasNet-A1

Retrain Top-1 Acc: 75.2. Result ID: 698d2c628b517eb4331f8d0b.

[Reporting paper](<https://www.sota2.com/research/paper/68fa67551e209131cea158f4.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa65da1e209131cea0cc69.md>)

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

Recorded evaluation configuration: {"FLOPS":"312M","No proxy":"False","Search Time (GPU hour)":"40000"}. This record does not assert independent reproduction.

### Proxyless-R (mobile)

Retrain Top-1 Acc: 74.6. Result ID: 698d2c628b517eb4331f8d0e.

[Reporting paper](<https://www.sota2.com/research/paper/68fa67551e209131cea158f4.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa65651e209131cea09fd7.md>)

[Implementation](<https://github.com/MIT-HAN-LAB/ProxylessNAS>)

Recorded evaluation configuration: {"FLOPS":"320M","No proxy":"True","Search Time (GPU hour)":"300+","Retrain Time (GPU hour)":"\>384"}. This record does not assert independent reproduction.

### Single Path One-Shot

Retrain Top-1 Acc: 74.3. Result ID: 698d2c628b517eb4331f8d0f.

[Reporting paper](<https://www.sota2.com/research/paper/68fa67551e209131cea158f4.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa67ed1e209131cea19308.md>)

Recorded evaluation configuration: {"FLOPS":"319M","No proxy":"True","Search Time (GPU hour)":"250","Retrain Time (GPU hour)":"384"}. This record does not assert independent reproduction.

### PNASNET

Retrain Top-1 Acc: 74.2. Result ID: 698d2c628b517eb4331f8d0a.

[Reporting paper](<https://www.sota2.com/research/paper/68fa67551e209131cea158f4.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa64a51e209131cea05d27.md>)

[Implementation](<http://github.com/tensorflow/models/>)

Recorded evaluation configuration: {"FLOPS":"588M","No proxy":"False","Search Time (GPU hour)":"5400"}. This record does not assert independent reproduction.

### NASNET-A(4@1056)

Retrain Top-1 Acc: 74. Result ID: 698d2c628b517eb4331f8d09.

[Reporting paper](<https://www.sota2.com/research/paper/68fa67551e209131cea158f4.md>)

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

Recorded evaluation configuration: {"FLOPS":"564M","No proxy":"False","Search Time (GPU hour)":"48000"}. This record does not assert independent reproduction.

### DARTS

Retrain Top-1 Acc: 73.3. Result ID: 698d2c628b517eb4331f8d0c.

[Reporting paper](<https://www.sota2.com/research/paper/68fa67551e209131cea158f4.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa65a91e209131cea0bb86.md>)

[Implementation](<https://github.com/quark0/darts>)

Recorded evaluation configuration: {"FLOPS":"574M","No proxy":"False","Search Time (GPU hour)":"24","Retrain Time (GPU hour)":"288"}. This record does not assert independent reproduction.

### SNAS

Retrain Top-1 Acc: 72.7. Result ID: 698d2c628b517eb4331f8d0d.

[Reporting paper](<https://www.sota2.com/research/paper/68fa67551e209131cea158f4.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa67541e209131cea1584d.md>)

[Implementation](<https://github.com/SNAS-Series/SNAS-Series>)

Recorded evaluation configuration: {"FLOPS":"522M","No proxy":"False","Search Time (GPU hour)":"36","Retrain Time (GPU hour)":"288"}. This record does not assert independent reproduction.

### ShuffleNet V2 (1.5x)

Retrain Top-1 Acc: 72.6. Result ID: 698d2c628b517eb4331f8d08.

[Reporting paper](<https://www.sota2.com/research/paper/68fa67551e209131cea158f4.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa64a61e209131cea05ddb.md>)

Recorded evaluation configuration: {"FLOPS":"299M","No proxy":"Manual","Search Time (GPU hour)":"Manual"}. This record does not assert independent reproduction.

### MobileNet V2 (1.0x)

Retrain Top-1 Acc: 72. Result ID: 698d2c628b517eb4331f8d07.

[Reporting paper](<https://www.sota2.com/research/paper/68fa67551e209131cea158f4.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa65811e209131cea0a9a1.md>)

[Implementation](<https://github.com/tensorflow/models/tree/master/research/slim/nets/mobilenet>)

Recorded evaluation configuration: {"FLOPS":"300M","No proxy":"Manual","Search Time (GPU hour)":"Manual"}. This record does not assert independent reproduction.

### MobileNet V1 (0.75x)

Retrain Top-1 Acc: 68.4. Result ID: 698d2c628b517eb4331f8d06.

[Reporting paper](<https://www.sota2.com/research/paper/68fa67551e209131cea158f4.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa634e1e209131ce9fdf55.md>)

Recorded evaluation configuration: {"FLOPS":"325M","No proxy":"Manual","Search Time (GPU hour)":"Manual"}. This record does not assert independent reproduction.

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