|
--- |
|
tags: |
|
- vision |
|
widget: |
|
- src: https://huggingface.co/OFA-Sys/chinese-clip-vit-base-patch16/resolve/main/festival.jpg |
|
candidate_labels: 灯笼, 鞭炮, 对联 |
|
example_title: festival |
|
- src: https://huggingface.co/datasets/mishig/sample_images/resolve/main/cat-dog-music.png |
|
candidate_labels: 音乐表演, 体育运动 |
|
example_title: cat & dog |
|
- src: https://huggingface.co/datasets/mishig/sample_images/resolve/main/football-match.jpg |
|
candidate_labels: 梅西, C罗, 马奎尔 |
|
example_title: football |
|
--- |
|
|
|
|
|
|
|
# Chinese-CLIP-ViT-Huge-Patch14 |
|
|
|
## Introduction |
|
This is the huge-version of the Chinese CLIP, with ViT-H/14 as the image encoder and RoBERTa-wwm-large as the text encoder. Chinese CLIP is a simple implementation of CLIP on a large-scale dataset of around 200 million Chinese image-text pairs. For more details, please refer to our technical report https://arxiv.org/abs/2211.01335 and our official github repo https://github.com/OFA-Sys/Chinese-CLIP (Welcome to star! 🔥🔥) |
|
|
|
## Use with the official API |
|
We provide a simple code snippet to show how to use the API of Chinese-CLIP to compute the image & text embeddings and similarities. |
|
|
|
```python |
|
from PIL import Image |
|
import requests |
|
from transformers import ChineseCLIPProcessor, ChineseCLIPModel |
|
|
|
model = ChineseCLIPModel.from_pretrained("OFA-Sys/chinese-clip-vit-huge-patch14") |
|
processor = ChineseCLIPProcessor.from_pretrained("OFA-Sys/chinese-clip-vit-huge-patch14") |
|
|
|
url = "https://clip-cn-beijing.oss-cn-beijing.aliyuncs.com/pokemon.jpeg" |
|
image = Image.open(requests.get(url, stream=True).raw) |
|
# Squirtle, Bulbasaur, Charmander, Pikachu in English |
|
texts = ["杰尼龟", "妙蛙种子", "小火龙", "皮卡丘"] |
|
|
|
# compute image feature |
|
inputs = processor(images=image, return_tensors="pt") |
|
image_features = model.get_image_features(**inputs) |
|
image_features = image_features / image_features.norm(p=2, dim=-1, keepdim=True) # normalize |
|
|
|
# compute text features |
|
inputs = processor(text=texts, padding=True, return_tensors="pt") |
|
text_features = model.get_text_features(**inputs) |
|
text_features = text_features / text_features.norm(p=2, dim=-1, keepdim=True) # normalize |
|
|
|
# compute image-text similarity scores |
|
inputs = processor(text=texts, images=image, return_tensors="pt", padding=True) |
|
outputs = model(**inputs) |
|
logits_per_image = outputs.logits_per_image # this is the image-text similarity score |
|
probs = logits_per_image.softmax(dim=1) # probs: [[1.1419e-02, 1.0478e-02, 5.2018e-04, 9.7758e-01]] |
|
``` |
|
|
|
However, if you are not satisfied with only using the API, feel free to check our github repo https://github.com/OFA-Sys/Chinese-CLIP for more details about training and inference. |
|
<br><br> |
|
|
|
## Results |
|
**MUGE Text-to-Image Retrieval**: |
|
<table border="1" width="100%"> |
|
<tr align="center"> |
|
<th>Setup</th><th colspan="4">Zero-shot</th><th colspan="4">Finetune</th> |
|
</tr> |
|
<tr align="center"> |
|
<td>Metric</td><td>R@1</td><td>R@5</td><td>R@10</td><td>MR</td><td>R@1</td><td>R@5</td><td>R@10</td><td>MR</td> |
|
</tr> |
|
<tr align="center"> |
|
<td width="120%">Wukong</td><td>42.7</td><td>69.0</td><td>78.0</td><td>63.2</td><td>52.7</td><td>77.9</td><td>85.6</td><td>72.1</td> |
|
</tr> |
|
<tr align="center"> |
|
<td width="120%">R2D2</td><td>49.5</td><td>75.7</td><td>83.2</td><td>69.5</td><td>60.1</td><td>82.9</td><td>89.4</td><td>77.5</td> |
|
</tr> |
|
<tr align="center"> |
|
<td width="120%">CN-CLIP</td><td>63.0</td><td>84.1</td><td>89.2</td><td>78.8</td><td>68.9</td><td>88.7</td><td>93.1</td><td>83.6</td> |
|
</tr> |
|
</table> |
|
<br> |
|
|
|
**Flickr30K-CN Retrieval**: |
|
<table border="1" width="120%"> |
|
<tr align="center"> |
|
<th>Task</th><th colspan="6">Text-to-Image</th><th colspan="6">Image-to-Text</th> |
|
</tr> |
|
<tr align="center"> |
|
<th>Setup</th><th colspan="3">Zero-shot</th><th colspan="3">Finetune</th><th colspan="3">Zero-shot</th><th colspan="3">Finetune</th> |
|
</tr> |
|
<tr align="center"> |
|
<td>Metric</td><td>R@1</td><td>R@5</td><td>R@10</td><td>R@1</td><td>R@5</td><td>R@10</td><td>R@1</td><td>R@5</td><td>R@10</td><td>R@1</td><td>R@5</td><td>R@10</td> |
|
</tr> |
|
<tr align="center"> |
|
<td width="120%">Wukong</td><td>51.7</td><td>78.9</td><td>86.3</td><td>77.4</td><td>94.5</td><td>97.0</td><td>76.1</td><td>94.8</td><td>97.5</td><td>92.7</td><td>99.1</td><td>99.6</td> |
|
</tr> |
|
<tr align="center"> |
|
<td width="120%">R2D2</td><td>60.9</td><td>86.8</td><td>92.7</td><td>84.4</td><td>96.7</td><td>98.4</td><td>77.6</td><td>96.7</td><td>98.9</td><td>95.6</td><td>99.8</td><td>100.0</td> |
|
</tr> |
|
<tr align="center"> |
|
<td width="120%">CN-CLIP</td><td>71.2</td><td>91.4</td><td>95.5</td><td>83.8</td><td>96.9</td><td>98.6</td><td>81.6</td><td>97.5</td><td>98.8</td><td>95.3</td><td>99.7</td><td>100.0</td> |
|
</tr> |
|
</table> |
|
<br> |
|
|
|
**COCO-CN Retrieval**: |
|
<table border="1" width="100%"> |
|
<tr align="center"> |
|
<th>Task</th><th colspan="6">Text-to-Image</th><th colspan="6">Image-to-Text</th> |
|
</tr> |
|
<tr align="center"> |
|
<th>Setup</th><th colspan="3">Zero-shot</th><th colspan="3">Finetune</th><th colspan="3">Zero-shot</th><th colspan="3">Finetune</th> |
|
</tr> |
|
<tr align="center"> |
|
<td>Metric</td><td>R@1</td><td>R@5</td><td>R@10</td><td>R@1</td><td>R@5</td><td>R@10</td><td>R@1</td><td>R@5</td><td>R@10</td><td>R@1</td><td>R@5</td><td>R@10</td> |
|
</tr> |
|
<tr align="center"> |
|
<td width="120%">Wukong</td><td>53.4</td><td>80.2</td><td>90.1</td><td>74.0</td><td>94.4</td><td>98.1</td><td>55.2</td><td>81.0</td><td>90.6</td><td>73.3</td><td>94.0</td><td>98.0</td> |
|
</tr> |
|
<tr align="center"> |
|
<td width="120%">R2D2</td><td>56.4</td><td>85.0</td><td>93.1</td><td>79.1</td><td>96.5</td><td>98.9</td><td>63.3</td><td>89.3</td><td>95.7</td><td>79.3</td><td>97.1</td><td>98.7</td> |
|
</tr> |
|
<tr align="center"> |
|
<td width="120%">CN-CLIP</td><td>69.2</td><td>89.9</td><td>96.1</td><td>81.5</td><td>96.9</td><td>99.1</td><td>63.0</td><td>86.6</td><td>92.9</td><td>83.5</td><td>97.3</td><td>99.2</td> |
|
</tr> |
|
</table> |
|
<br> |
|
|
|
**Zero-shot Image Classification**: |
|
<table border="1" width="100%"> |
|
<tr align="center"> |
|
<th>Task</th><th>CIFAR10</th><th>CIFAR100</th><th>DTD</th><th>EuroSAT</th><th>FER</th><th>FGVC</th><th>KITTI</th><th>MNIST</th><th>PC</th><th>VOC</th> |
|
</tr> |
|
<tr align="center"> |
|
<td width="150%">GIT</td><td>88.5</td><td>61.1</td><td>42.9</td><td>43.4</td><td>41.4</td><td>6.7</td><td>22.1</td><td>68.9</td><td>50.0</td><td>80.2</td> |
|
</tr> |
|
<tr align="center"> |
|
<td width="150%">ALIGN</td><td>94.9</td><td>76.8</td><td>66.1</td><td>52.1</td><td>50.8</td><td>25.0</td><td>41.2</td><td>74.0</td><td>55.2</td><td>83.0</td> |
|
</tr> |
|
<tr align="center"> |
|
<td width="150%">CLIP</td><td>94.9</td><td>77.0</td><td>56.0</td><td>63.0</td><td>48.3</td><td>33.3</td><td>11.5</td><td>79.0</td><td>62.3</td><td>84.0</td> |
|
</tr> |
|
<tr align="center"> |
|
<td width="150%">Wukong</td><td>95.4</td><td>77.1</td><td>40.9</td><td>50.3</td><td>-</td><td>-</td><td>-</td><td>-</td><td>-</td><td>-</td> |
|
</tr> |
|
<tr align="center"> |
|
<td width="150%">CN-CLIP</td><td>96.0</td><td>79.7</td><td>51.2</td><td>52.0</td><td>55.1</td><td>26.2</td><td>49.9</td><td>79.4</td><td>63.5</td><td>84.9</td> |
|
</tr> |
|
</table> |
|
<br> |
|
|
|
## Citation |
|
If you find Chinese CLIP helpful, feel free to cite our paper. Thanks for your support! |
|
|
|
``` |
|
@article{chinese-clip, |
|
title={Chinese CLIP: Contrastive Vision-Language Pretraining in Chinese}, |
|
author={Yang, An and Pan, Junshu and Lin, Junyang and Men, Rui and Zhang, Yichang and Zhou, Jingren and Zhou, Chang}, |
|
journal={arXiv preprint arXiv:2211.01335}, |
|
year={2022} |
|
} |
|
``` |
|
<br> |
