# Action Recognition on HMDB51 (Mean Accuracy)

Source: [Action Recognition on HMDB51 (Mean Accuracy)](<https://www.sota2.com/research/sota/action-recognition-on-hmdb51-mean-accuracy>)

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

Data updated: 2026-02-26T05:51:52.581Z

Evaluation of video action recognition performance using the HMDB51 dataset, measured by Mean Accuracy.

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 | Action Recognition |
| Dataset (including any stated split) | HMDB51 |
| Metric | Mean Accuracy |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 61 of 61 |
| 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

- [Mean Accuracy](<https://www.sota2.com/research/sota/action-recognition-on-hmdb51-mean-accuracy.md?metric=Mean%20Accuracy>): Higher is better; unit: PERCENT.

## Recorded results

| Method | Mean Accuracy | Result date | Report paper |
| --- | --- | --- | --- |
| Ours ViT-L | 81.3 | Not recorded | 68fa6ee71e209131cea42c8a |
| Ours ViT-L (336↑) | 81.3 | Not recorded | 68fa6ee71e209131cea42c8a |
| TDN | 76.4 | Not recorded | 68fa6ee71e209131cea42c8a |
| TDN | 76.3 | Not recorded | 68fa69b11e209131cea23747 |
| S3D | 75.9 | Not recorded | 68fa69b11e209131cea23747 |
| S3D-G | 75.9 | Not recorded | 68fa6ee71e209131cea42c8a |
| MVFNet | 75.7 | Not recorded | 68fa6ee71e209131cea42c8a |
| I3D | 74.8 | Not recorded | 68fa69b11e209131cea23747 |
| I3D | 74.8 | Not recorded | 68fa6ee71e209131cea42c8a |
| R(2+1)D | 74.5 | Not recorded | 68fa69b11e209131cea23747 |
| R(2+1)D | 74.5 | Not recorded | 68fa6ee71e209131cea42c8a |
| TSM | 73.5 | Not recorded | 68fa6ee71e209131cea42c8a |
| TEA | 73.3 | Not recorded | 68fa69b11e209131cea23747 |
| TSM | 73.2 | Not recorded | 68fa69b11e209131cea23747 |
| STM | 72.2 | Not recorded | 68fa69b11e209131cea23747 |
| STM | 72.2 | Not recorded | 68fa6ee71e209131cea42c8a |
| TEINet | 72.1 | Not recorded | 68fa6ee71e209131cea42c8a |
| ARTNet | 70.9 | Not recorded | 68fa69b11e209131cea23747 |
| ARTNet | 70.9 | Not recorded | 68fa6ee71e209131cea42c8a |
| GA2-CLIP | 70.8 | Not recorded | 692d2108d6b0dd334293161e |
| ViLT-CLIP | 69.6 | Not recorded | 692d2108d6b0dd334293161e |
| GA2-CLIP | 68.1 | Not recorded | 692d2108d6b0dd334293161e |
| ViFi-CLIP | 67.4 | Not recorded | 692d2108d6b0dd334293161e |
| ViLT-CLIP | 66.9 | Not recorded | 692d2108d6b0dd334293161e |
| ViFi-CLIP | 66.8 | Not recorded | 692d2108d6b0dd334293161e |
| CLIP text-FT | 65 | Not recorded | 692d2108d6b0dd334293161e |
| ViFi-CLIP | 64.8 | Not recorded | 692d2108d6b0dd334293161e |
| ViFi-CLIP | 64.5 | Not recorded | 692d2108d6b0dd334293161e |
| GA2-CLIP | 64.3 | Not recorded | 692d2108d6b0dd334293161e |
| X-CLIP | 64 | Not recorded | 692d2108d6b0dd334293161e |
| CLIP text-FT | 63.1 | Not recorded | 692d2108d6b0dd334293161e |
| X-CLIP | 62.8 | Not recorded | 692d2108d6b0dd334293161e |
| ViFi-CLIP | 62.7 | Not recorded | 692d2108d6b0dd334293161e |
| A5 | 62.4 | Not recorded | 692d2108d6b0dd334293161e |
| CLIP image-FT | 62 | Not recorded | 692d2108d6b0dd334293161e |
| ViLT-CLIP | 61.9 | Not recorded | 692d2108d6b0dd334293161e |
| GA2-CLIP | 61.9 | Not recorded | 692d2108d6b0dd334293161e |
| CLIP text-FT | 61.6 | Not recorded | 692d2108d6b0dd334293161e |
| ViLT-CLIP | 60.6 | Not recorded | 692d2108d6b0dd334293161e |
| ViFi-CLIP | 60 | Not recorded | 692d2108d6b0dd334293161e |
| ActionCLIP | 59.1 | Not recorded | 692d2108d6b0dd334293161e |
| ActionCLIP | 57.9 | Not recorded | 692d2108d6b0dd334293161e |
| CLIP image-FT | 57.8 | Not recorded | 692d2108d6b0dd334293161e |
| ViFi-CLIP | 57.8 | Not recorded | 692d2108d6b0dd334293161e |
| ActionCLIP | 57.3 | Not recorded | 692d2108d6b0dd334293161e |
| X-CLIP | 57.3 | Not recorded | 692d2108d6b0dd334293161e |
| ViFi-CLIP | 57.2 | Not recorded | 692d2108d6b0dd334293161e |
| A5 | 56 | Not recorded | 692d2108d6b0dd334293161e |
| C3D | 54.9 | Not recorded | 68fa69b11e209131cea23747 |
| CLIP image-FT | 54.9 | Not recorded | 692d2108d6b0dd334293161e |
| CLIP text-FT | 54.5 | Not recorded | 692d2108d6b0dd334293161e |
| TSN | 53.7 | Not recorded | 68fa69b11e209131cea23747 |
| X-CLIP | 53 | Not recorded | 692d2108d6b0dd334293161e |
| A5 | 50.7 | Not recorded | 692d2108d6b0dd334293161e |
| CLIP image-FT | 49.6 | Not recorded | 692d2108d6b0dd334293161e |
| ActionCLIP | 47.5 | Not recorded | 692d2108d6b0dd334293161e |
| Vanilla CLIP | 41.9 | Not recorded | 692d2108d6b0dd334293161e |
| Vanilla CLIP | 41.9 | Not recorded | 692d2108d6b0dd334293161e |
| Vanilla CLIP | 41.9 | Not recorded | 692d2108d6b0dd334293161e |
| Vanilla CLIP | 41.9 | Not recorded | 692d2108d6b0dd334293161e |
| A5 | 39.7 | Not recorded | 692d2108d6b0dd334293161e |

### Ours ViT-L

Mean Accuracy: 81.3. Result ID: 699082bd3a6c314849011673.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6ee71e209131cea42c8a.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6ee71e209131cea42c8a.md>)

[Implementation](<https://github.com/whwu95/Text4Vis>)

Recorded evaluation configuration: {"Modality":"RGB","Pre-training":"Kinetics","Transfer Protocol":"Transfer Learning","Backbone":"ViT-L"}. This record does not assert independent reproduction.

### Ours ViT-L (336↑)

Mean Accuracy: 81.3. Result ID: 699082bd3a6c314849011674.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6ee71e209131cea42c8a.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6ee71e209131cea42c8a.md>)

[Implementation](<https://github.com/whwu95/Text4Vis>)

Recorded evaluation configuration: {"Modality":"RGB","Pre-training":"Kinetics","Transfer Protocol":"Transfer Learning","Backbone":"ViT-L","Resolution":"336"}. This record does not assert independent reproduction.

### TDN

Mean Accuracy: 76.4. Result ID: 699082bd3a6c314849011672.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6ee71e209131cea42c8a.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa69b11e209131cea23747.md>)

[Implementation](<https://github.com/MCG-NJU/TDN>)

Recorded evaluation configuration: {"Modality":"RGB","Pre-training":"Kinetics","Transfer Protocol":"Transfer Learning"}. This record does not assert independent reproduction.

### TDN

Mean Accuracy: 76.3. Result ID: 698dd424ffc3901de098e989.

[Reporting paper](<https://www.sota2.com/research/paper/68fa69b11e209131cea23747.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa69b11e209131cea23747.md>)

[Implementation](<https://github.com/MCG-NJU/TDN>)

Recorded evaluation configuration: {"Pretrain":"ImageNet + Kinetics","Backbone":"ResNet50"}. This record does not assert independent reproduction.

### S3D

Mean Accuracy: 75.9. Result ID: 698dd424ffc3901de098e984.

[Reporting paper](<https://www.sota2.com/research/paper/68fa69b11e209131cea23747.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa64a51e209131cea05d34.md>)

Recorded evaluation configuration: {"Pretrain":"ImageNet+Kinetics","Backbone":"Inception V2"}. This record does not assert independent reproduction.

### S3D-G

Mean Accuracy: 75.9. Result ID: 699082bd3a6c31484901166d.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6ee71e209131cea42c8a.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa64a51e209131cea05d34.md>)

Recorded evaluation configuration: {"Modality":"RGB","Pre-training":"Kinetics","Transfer Protocol":"Transfer Learning"}. This record does not assert independent reproduction.

### MVFNet

Mean Accuracy: 75.7. Result ID: 699082bd3a6c314849011671.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6ee71e209131cea42c8a.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa69101e209131cea1ff08.md>)

[Implementation](<https://github.com/whwu95/MVFNet>)

Recorded evaluation configuration: {"Modality":"RGB","Pre-training":"Kinetics","Transfer Protocol":"Transfer Learning"}. This record does not assert independent reproduction.

### I3D

Mean Accuracy: 74.8. Result ID: 698dd424ffc3901de098e982.

[Reporting paper](<https://www.sota2.com/research/paper/68fa69b11e209131cea23747.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa64051e209131cea020a8.md>)

Recorded evaluation configuration: {"Pretrain":"ImageNet+Kinetics","Backbone":"Inception V2"}. This record does not assert independent reproduction.

### I3D

Mean Accuracy: 74.8. Result ID: 699082bd3a6c31484901166b.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6ee71e209131cea42c8a.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa64051e209131cea020a8.md>)

Recorded evaluation configuration: {"Modality":"RGB","Pre-training":"Kinetics","Transfer Protocol":"Transfer Learning"}. This record does not assert independent reproduction.

### R(2+1)D

Mean Accuracy: 74.5. Result ID: 698dd424ffc3901de098e985.

[Reporting paper](<https://www.sota2.com/research/paper/68fa69b11e209131cea23747.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa64461e209131cea03a61.md>)

Recorded evaluation configuration: {"Pretrain":"Kinetics","Backbone":"ResNet34"}. This record does not assert independent reproduction.

### R(2+1)D

Mean Accuracy: 74.5. Result ID: 699082bd3a6c31484901166c.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6ee71e209131cea42c8a.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa64461e209131cea03a61.md>)

Recorded evaluation configuration: {"Modality":"RGB","Pre-training":"Kinetics","Transfer Protocol":"Transfer Learning"}. This record does not assert independent reproduction.

### TSM

Mean Accuracy: 73.5. Result ID: 699082bd3a6c31484901166e.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6ee71e209131cea42c8a.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa66381e209131cea0ee93.md>)

[Implementation](<https://github.com/mit-han-lab/temporal-shift-module>)

Recorded evaluation configuration: {"Modality":"RGB","Pre-training":"Kinetics","Transfer Protocol":"Transfer Learning"}. This record does not assert independent reproduction.

### TEA

Mean Accuracy: 73.3. Result ID: 698dd424ffc3901de098e988.

[Reporting paper](<https://www.sota2.com/research/paper/68fa69b11e209131cea23747.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa67571e209131cea15a94.md>)

Recorded evaluation configuration: {"Pretrain":"ImageNet + Kinetics","Backbone":"ResNet50"}. This record does not assert independent reproduction.

### TSM

Mean Accuracy: 73.2. Result ID: 698dd424ffc3901de098e986.

[Reporting paper](<https://www.sota2.com/research/paper/68fa69b11e209131cea23747.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa66381e209131cea0ee93.md>)

[Implementation](<https://github.com/mit-han-lab/temporal-shift-module>)

Recorded evaluation configuration: {"Pretrain":"Kinetics","Backbone":"ResNet50"}. This record does not assert independent reproduction.

### STM

Mean Accuracy: 72.2. Result ID: 698dd424ffc3901de098e987.

[Reporting paper](<https://www.sota2.com/research/paper/68fa69b11e209131cea23747.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa66331e209131cea0eb46.md>)

Recorded evaluation configuration: {"Pretrain":"ImageNet + Kinetics","Backbone":"ResNet50"}. This record does not assert independent reproduction.

### STM

Mean Accuracy: 72.2. Result ID: 699082bd3a6c31484901166f.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6ee71e209131cea42c8a.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa66331e209131cea0eb46.md>)

Recorded evaluation configuration: {"Modality":"RGB","Pre-training":"Kinetics","Transfer Protocol":"Transfer Learning"}. This record does not assert independent reproduction.

### TEINet

Mean Accuracy: 72.1. Result ID: 699082bd3a6c314849011670.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6ee71e209131cea42c8a.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa66b51e209131cea11fd9.md>)

Recorded evaluation configuration: {"Modality":"RGB","Pre-training":"Kinetics","Transfer Protocol":"Transfer Learning"}. This record does not assert independent reproduction.

### ARTNet

Mean Accuracy: 70.9. Result ID: 698dd424ffc3901de098e983.

[Reporting paper](<https://www.sota2.com/research/paper/68fa69b11e209131cea23747.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa64511e209131cea0409e.md>)

[Implementation](<https://github.com/wanglimin/ARTNet>)

Recorded evaluation configuration: {"Pretrain":"Kinetics","Backbone":"ResNet18"}. This record does not assert independent reproduction.

### ARTNet

Mean Accuracy: 70.9. Result ID: 699082bd3a6c31484901166a.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6ee71e209131cea42c8a.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa64511e209131cea0409e.md>)

[Implementation](<https://github.com/wanglimin/ARTNet>)

Recorded evaluation configuration: {"Modality":"RGB","Pre-training":"Kinetics","Transfer Protocol":"Transfer Learning"}. This record does not assert independent reproduction.

### GA2-CLIP

Mean Accuracy: 70.8. Result ID: 6995e71d3badfaffe6ef52a1.

[Reporting paper](<https://www.sota2.com/research/paper/692d2108d6b0dd334293161e.md>)

[Method origin paper](<https://www.sota2.com/research/paper/692d2108d6b0dd334293161e.md>)

Recorded evaluation configuration: {"K":16,"Adaptation Protocol":"Prompt tuning pre-trained image VL models"}. This record does not assert independent reproduction.

### ViLT-CLIP

Mean Accuracy: 69.6. Result ID: 6995e71d3badfaffe6ef529d.

[Reporting paper](<https://www.sota2.com/research/paper/692d2108d6b0dd334293161e.md>)

Recorded evaluation configuration: {"K":16,"Adaptation Protocol":"Prompt tuning pre-trained image VL models"}. This record does not assert independent reproduction.

### GA2-CLIP

Mean Accuracy: 68.1. Result ID: 6995e71d3badfaffe6ef52a0.

[Reporting paper](<https://www.sota2.com/research/paper/692d2108d6b0dd334293161e.md>)

[Method origin paper](<https://www.sota2.com/research/paper/692d2108d6b0dd334293161e.md>)

Recorded evaluation configuration: {"K":8,"Adaptation Protocol":"Prompt tuning pre-trained image VL models"}. This record does not assert independent reproduction.

### ViFi-CLIP

Mean Accuracy: 67.4. Result ID: 6995e71d3badfaffe6ef5299.

[Reporting paper](<https://www.sota2.com/research/paper/692d2108d6b0dd334293161e.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6ee81e209131cea42cef.md>)

[Implementation](<https://github.com/muzairkhattak/ViFi-CLIP>)

Recorded evaluation configuration: {"K":16,"Adaptation Protocol":"Prompt tuning pre-trained image VL models"}. This record does not assert independent reproduction.

### ViLT-CLIP

Mean Accuracy: 66.9. Result ID: 6995e71d3badfaffe6ef529c.

[Reporting paper](<https://www.sota2.com/research/paper/692d2108d6b0dd334293161e.md>)

Recorded evaluation configuration: {"K":8,"Adaptation Protocol":"Prompt tuning pre-trained image VL models"}. This record does not assert independent reproduction.

### ViFi-CLIP

Mean Accuracy: 66.8. Result ID: 6995e71d3badfaffe6ef5295.

[Reporting paper](<https://www.sota2.com/research/paper/692d2108d6b0dd334293161e.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6ee81e209131cea42cef.md>)

[Implementation](<https://github.com/muzairkhattak/ViFi-CLIP>)

Recorded evaluation configuration: {"K":16,"Adaptation Protocol":"Tuning pre-trained image VL models"}. This record does not assert independent reproduction.

### CLIP text-FT

Mean Accuracy: 65. Result ID: 6995e71d3badfaffe6ef5291.

[Reporting paper](<https://www.sota2.com/research/paper/692d2108d6b0dd334293161e.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6ee81e209131cea42cef.md>)

[Implementation](<https://github.com/muzairkhattak/ViFi-CLIP>)

Recorded evaluation configuration: {"K":16,"Adaptation Protocol":"Tuning pre-trained image VL models"}. This record does not assert independent reproduction.

### ViFi-CLIP

Mean Accuracy: 64.8. Result ID: 6995e71d3badfaffe6ef5298.

[Reporting paper](<https://www.sota2.com/research/paper/692d2108d6b0dd334293161e.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6ee81e209131cea42cef.md>)

[Implementation](<https://github.com/muzairkhattak/ViFi-CLIP>)

Recorded evaluation configuration: {"K":8,"Adaptation Protocol":"Prompt tuning pre-trained image VL models"}. This record does not assert independent reproduction.

### ViFi-CLIP

Mean Accuracy: 64.5. Result ID: 6995e71d3badfaffe6ef5294.

[Reporting paper](<https://www.sota2.com/research/paper/692d2108d6b0dd334293161e.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6ee81e209131cea42cef.md>)

[Implementation](<https://github.com/muzairkhattak/ViFi-CLIP>)

Recorded evaluation configuration: {"K":8,"Adaptation Protocol":"Tuning pre-trained image VL models"}. This record does not assert independent reproduction.

### GA2-CLIP

Mean Accuracy: 64.3. Result ID: 6995e71d3badfaffe6ef529f.

[Reporting paper](<https://www.sota2.com/research/paper/692d2108d6b0dd334293161e.md>)

[Method origin paper](<https://www.sota2.com/research/paper/692d2108d6b0dd334293161e.md>)

Recorded evaluation configuration: {"K":4,"Adaptation Protocol":"Prompt tuning pre-trained image VL models"}. This record does not assert independent reproduction.

### X-CLIP

Mean Accuracy: 64. Result ID: 6995e71d3badfaffe6ef5285.

[Reporting paper](<https://www.sota2.com/research/paper/692d2108d6b0dd334293161e.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6d331e209131cea385df.md>)

[Implementation](<https://www.sota2.com/aka.ms/X-CLIP>)

Recorded evaluation configuration: {"K":16,"Adaptation Protocol":"Adapting pre-trained image VL models"}. This record does not assert independent reproduction.

### CLIP text-FT

Mean Accuracy: 63.1. Result ID: 6995e71d3badfaffe6ef5290.

[Reporting paper](<https://www.sota2.com/research/paper/692d2108d6b0dd334293161e.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6ee81e209131cea42cef.md>)

[Implementation](<https://github.com/muzairkhattak/ViFi-CLIP>)

Recorded evaluation configuration: {"K":8,"Adaptation Protocol":"Tuning pre-trained image VL models"}. This record does not assert independent reproduction.

### X-CLIP

Mean Accuracy: 62.8. Result ID: 6995e71d3badfaffe6ef5284.

[Reporting paper](<https://www.sota2.com/research/paper/692d2108d6b0dd334293161e.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6d331e209131cea385df.md>)

[Implementation](<https://www.sota2.com/aka.ms/X-CLIP>)

Recorded evaluation configuration: {"K":8,"Adaptation Protocol":"Adapting pre-trained image VL models"}. This record does not assert independent reproduction.

### ViFi-CLIP

Mean Accuracy: 62.7. Result ID: 6995e71d3badfaffe6ef5293.

[Reporting paper](<https://www.sota2.com/research/paper/692d2108d6b0dd334293161e.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6ee81e209131cea42cef.md>)

[Implementation](<https://github.com/muzairkhattak/ViFi-CLIP>)

Recorded evaluation configuration: {"K":4,"Adaptation Protocol":"Tuning pre-trained image VL models"}. This record does not assert independent reproduction.

### A5

Mean Accuracy: 62.4. Result ID: 6995e71d3badfaffe6ef5289.

[Reporting paper](<https://www.sota2.com/research/paper/692d2108d6b0dd334293161e.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6d041e209131cea375a1.md>)

Recorded evaluation configuration: {"K":16,"Adaptation Protocol":"Adapting pre-trained image VL models"}. This record does not assert independent reproduction.

### CLIP image-FT

Mean Accuracy: 62. Result ID: 6995e71d3badfaffe6ef528d.

[Reporting paper](<https://www.sota2.com/research/paper/692d2108d6b0dd334293161e.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6ee81e209131cea42cef.md>)

[Implementation](<https://github.com/muzairkhattak/ViFi-CLIP>)

Recorded evaluation configuration: {"K":16,"Adaptation Protocol":"Tuning pre-trained image VL models"}. This record does not assert independent reproduction.

### ViLT-CLIP

Mean Accuracy: 61.9. Result ID: 6995e71d3badfaffe6ef529b.

[Reporting paper](<https://www.sota2.com/research/paper/692d2108d6b0dd334293161e.md>)

Recorded evaluation configuration: {"K":4,"Adaptation Protocol":"Prompt tuning pre-trained image VL models"}. This record does not assert independent reproduction.

### GA2-CLIP

Mean Accuracy: 61.9. Result ID: 6995e71d3badfaffe6ef529e.

[Reporting paper](<https://www.sota2.com/research/paper/692d2108d6b0dd334293161e.md>)

[Method origin paper](<https://www.sota2.com/research/paper/692d2108d6b0dd334293161e.md>)

Recorded evaluation configuration: {"K":2,"Adaptation Protocol":"Prompt tuning pre-trained image VL models"}. This record does not assert independent reproduction.

### CLIP text-FT

Mean Accuracy: 61.6. Result ID: 6995e71d3badfaffe6ef528f.

[Reporting paper](<https://www.sota2.com/research/paper/692d2108d6b0dd334293161e.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6ee81e209131cea42cef.md>)

[Implementation](<https://github.com/muzairkhattak/ViFi-CLIP>)

Recorded evaluation configuration: {"K":4,"Adaptation Protocol":"Tuning pre-trained image VL models"}. This record does not assert independent reproduction.

### ViLT-CLIP

Mean Accuracy: 60.6. Result ID: 6995e71d3badfaffe6ef529a.

[Reporting paper](<https://www.sota2.com/research/paper/692d2108d6b0dd334293161e.md>)

Recorded evaluation configuration: {"K":2,"Adaptation Protocol":"Prompt tuning pre-trained image VL models"}. This record does not assert independent reproduction.

### ViFi-CLIP

Mean Accuracy: 60. Result ID: 6995e71d3badfaffe6ef5297.

[Reporting paper](<https://www.sota2.com/research/paper/692d2108d6b0dd334293161e.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6ee81e209131cea42cef.md>)

[Implementation](<https://github.com/muzairkhattak/ViFi-CLIP>)

Recorded evaluation configuration: {"K":4,"Adaptation Protocol":"Prompt tuning pre-trained image VL models"}. This record does not assert independent reproduction.

### ActionCLIP

Mean Accuracy: 59.1. Result ID: 6995e71d3badfaffe6ef5281.

[Reporting paper](<https://www.sota2.com/research/paper/692d2108d6b0dd334293161e.md>)

Recorded evaluation configuration: {"K":16,"Adaptation Protocol":"Adapting pre-trained image VL models"}. This record does not assert independent reproduction.

### ActionCLIP

Mean Accuracy: 57.9. Result ID: 6995e71d3badfaffe6ef527f.

[Reporting paper](<https://www.sota2.com/research/paper/692d2108d6b0dd334293161e.md>)

Recorded evaluation configuration: {"K":4,"Adaptation Protocol":"Adapting pre-trained image VL models"}. This record does not assert independent reproduction.

### CLIP image-FT

Mean Accuracy: 57.8. Result ID: 6995e71d3badfaffe6ef528c.

[Reporting paper](<https://www.sota2.com/research/paper/692d2108d6b0dd334293161e.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6ee81e209131cea42cef.md>)

[Implementation](<https://github.com/muzairkhattak/ViFi-CLIP>)

Recorded evaluation configuration: {"K":8,"Adaptation Protocol":"Tuning pre-trained image VL models"}. This record does not assert independent reproduction.

### ViFi-CLIP

Mean Accuracy: 57.8. Result ID: 6995e71d3badfaffe6ef5296.

[Reporting paper](<https://www.sota2.com/research/paper/692d2108d6b0dd334293161e.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6ee81e209131cea42cef.md>)

[Implementation](<https://github.com/muzairkhattak/ViFi-CLIP>)

Recorded evaluation configuration: {"K":2,"Adaptation Protocol":"Prompt tuning pre-trained image VL models"}. This record does not assert independent reproduction.

### ActionCLIP

Mean Accuracy: 57.3. Result ID: 6995e71d3badfaffe6ef5280.

[Reporting paper](<https://www.sota2.com/research/paper/692d2108d6b0dd334293161e.md>)

Recorded evaluation configuration: {"K":8,"Adaptation Protocol":"Adapting pre-trained image VL models"}. This record does not assert independent reproduction.

### X-CLIP

Mean Accuracy: 57.3. Result ID: 6995e71d3badfaffe6ef5283.

[Reporting paper](<https://www.sota2.com/research/paper/692d2108d6b0dd334293161e.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6d331e209131cea385df.md>)

[Implementation](<https://www.sota2.com/aka.ms/X-CLIP>)

Recorded evaluation configuration: {"K":4,"Adaptation Protocol":"Adapting pre-trained image VL models"}. This record does not assert independent reproduction.

### ViFi-CLIP

Mean Accuracy: 57.2. Result ID: 6995e71d3badfaffe6ef5292.

[Reporting paper](<https://www.sota2.com/research/paper/692d2108d6b0dd334293161e.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6ee81e209131cea42cef.md>)

[Implementation](<https://github.com/muzairkhattak/ViFi-CLIP>)

Recorded evaluation configuration: {"K":2,"Adaptation Protocol":"Tuning pre-trained image VL models"}. This record does not assert independent reproduction.

### A5

Mean Accuracy: 56. Result ID: 6995e71d3badfaffe6ef5288.

[Reporting paper](<https://www.sota2.com/research/paper/692d2108d6b0dd334293161e.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6d041e209131cea375a1.md>)

Recorded evaluation configuration: {"K":8,"Adaptation Protocol":"Adapting pre-trained image VL models"}. This record does not assert independent reproduction.

### C3D

Mean Accuracy: 54.9. Result ID: 698dd424ffc3901de098e981.

[Reporting paper](<https://www.sota2.com/research/paper/68fa69b11e209131cea23747.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa62171e209131ce9f77be.md>)

[Implementation](<http://vlg.cs.dartmouth.edu/c3d>)

Recorded evaluation configuration: {"Pretrain":"Sports-1M","Backbone":"ResNet18"}. This record does not assert independent reproduction.

### CLIP image-FT

Mean Accuracy: 54.9. Result ID: 6995e71d3badfaffe6ef528b.

[Reporting paper](<https://www.sota2.com/research/paper/692d2108d6b0dd334293161e.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6ee81e209131cea42cef.md>)

[Implementation](<https://github.com/muzairkhattak/ViFi-CLIP>)

Recorded evaluation configuration: {"K":4,"Adaptation Protocol":"Tuning pre-trained image VL models"}. This record does not assert independent reproduction.

### CLIP text-FT

Mean Accuracy: 54.5. Result ID: 6995e71d3badfaffe6ef528e.

[Reporting paper](<https://www.sota2.com/research/paper/692d2108d6b0dd334293161e.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6ee81e209131cea42cef.md>)

[Implementation](<https://github.com/muzairkhattak/ViFi-CLIP>)

Recorded evaluation configuration: {"K":2,"Adaptation Protocol":"Tuning pre-trained image VL models"}. This record does not assert independent reproduction.

### TSN

Mean Accuracy: 53.7. Result ID: 698dd424ffc3901de098e980.

[Reporting paper](<https://www.sota2.com/research/paper/68fa69b11e209131cea23747.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa62a51e209131ce9fa8d4.md>)

[Implementation](<https://github.com/yjxiong/temporal-segment-networks>)

Recorded evaluation configuration: {"Pretrain":"ImageNet","Backbone":"Inception V2"}. This record does not assert independent reproduction.

### X-CLIP

Mean Accuracy: 53. Result ID: 6995e71d3badfaffe6ef5282.

[Reporting paper](<https://www.sota2.com/research/paper/692d2108d6b0dd334293161e.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6d331e209131cea385df.md>)

[Implementation](<https://www.sota2.com/aka.ms/X-CLIP>)

Recorded evaluation configuration: {"K":2,"Adaptation Protocol":"Adapting pre-trained image VL models"}. This record does not assert independent reproduction.

### A5

Mean Accuracy: 50.7. Result ID: 6995e71d3badfaffe6ef5287.

[Reporting paper](<https://www.sota2.com/research/paper/692d2108d6b0dd334293161e.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6d041e209131cea375a1.md>)

Recorded evaluation configuration: {"K":4,"Adaptation Protocol":"Adapting pre-trained image VL models"}. This record does not assert independent reproduction.

### CLIP image-FT

Mean Accuracy: 49.6. Result ID: 6995e71d3badfaffe6ef528a.

[Reporting paper](<https://www.sota2.com/research/paper/692d2108d6b0dd334293161e.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6ee81e209131cea42cef.md>)

[Implementation](<https://github.com/muzairkhattak/ViFi-CLIP>)

Recorded evaluation configuration: {"K":2,"Adaptation Protocol":"Tuning pre-trained image VL models"}. This record does not assert independent reproduction.

### ActionCLIP

Mean Accuracy: 47.5. Result ID: 6995e71d3badfaffe6ef527e.

[Reporting paper](<https://www.sota2.com/research/paper/692d2108d6b0dd334293161e.md>)

Recorded evaluation configuration: {"K":2,"Adaptation Protocol":"Adapting pre-trained image VL models"}. This record does not assert independent reproduction.

### Vanilla CLIP

Mean Accuracy: 41.9. Result ID: 6995e71d3badfaffe6ef527a.

[Reporting paper](<https://www.sota2.com/research/paper/692d2108d6b0dd334293161e.md>)

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

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

Recorded evaluation configuration: {"K":2,"Adaptation Protocol":"Adapting pre-trained image VL models"}. This record does not assert independent reproduction.

### Vanilla CLIP

Mean Accuracy: 41.9. Result ID: 6995e71d3badfaffe6ef527b.

[Reporting paper](<https://www.sota2.com/research/paper/692d2108d6b0dd334293161e.md>)

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

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

Recorded evaluation configuration: {"K":4,"Adaptation Protocol":"Adapting pre-trained image VL models"}. This record does not assert independent reproduction.

### Vanilla CLIP

Mean Accuracy: 41.9. Result ID: 6995e71d3badfaffe6ef527c.

[Reporting paper](<https://www.sota2.com/research/paper/692d2108d6b0dd334293161e.md>)

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

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

Recorded evaluation configuration: {"K":8,"Adaptation Protocol":"Adapting pre-trained image VL models"}. This record does not assert independent reproduction.

### Vanilla CLIP

Mean Accuracy: 41.9. Result ID: 6995e71d3badfaffe6ef527d.

[Reporting paper](<https://www.sota2.com/research/paper/692d2108d6b0dd334293161e.md>)

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

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

Recorded evaluation configuration: {"K":16,"Adaptation Protocol":"Adapting pre-trained image VL models"}. This record does not assert independent reproduction.

### A5

Mean Accuracy: 39.7. Result ID: 6995e71d3badfaffe6ef5286.

[Reporting paper](<https://www.sota2.com/research/paper/692d2108d6b0dd334293161e.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6d041e209131cea375a1.md>)

Recorded evaluation configuration: {"K":2,"Adaptation Protocol":"Adapting pre-trained image VL models"}. This record does not assert independent reproduction.

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