Taiyi-CLIP-Roberta-large-326M-Chinese

• Github: Fengshenbang-LM
• Docs: Fengshenbang-Docs

简介 Brief Introduction

首个开源的中文CLIP模型，1.23亿图文对上进行预训练的文本端RoBERTa-large。

The first open source Chinese CLIP, pre-training on 123M image-text pairs, the text encoder: RoBERTa-large.

模型分类 Model Taxonomy

模型信息 Model Information

我们遵循CLIP的实验设置，以获得强大的视觉-语言表征。在训练中文版的CLIP时，我们使用chinese-roberta-wwm-large作为语言的编码器，并将CLIP中的ViT-L-14应用于视觉的编码器。为了快速且稳定地进行预训练，我们冻结了视觉编码器并且只微调语言编码器。此外，我们将Noah-Wukong数据集(100M)和Zero数据集(23M)用作预训练的数据集。我们先在悟空数据集上预训练了10轮，然后接着在悟空数据集和zero数据集上预训练12轮， 在A100x16上训练了7天。据我们所知，我们的Taiyi-CLIP是目前Huggingface社区中首个的开源中文CLIP。

We follow the experimental setup of CLIP to obtain powerful visual-language intelligence. To obtain the CLIP for Chinese, we employ chinese-roberta-wwm-large for the language encoder, and apply the ViT-L-14 in CLIP for the vision encoder. We freeze the vision encoder and tune the language encoder to speed up and stabilize the pre-training process. Moreover, we apply Noah-Wukong dataset (100M) and Zero dataset (23M) as the pre-training datasets. The model was first trained 10 epochs on wukong and then train another 12 epochs on wukong and zero, which takes about 14 days to train on A100x16. To the best of our knowledge, our TaiyiCLIP is currently the only open-sourced Chinese CLIP in the huggingface community.

下游效果 Performance

Zero-Shot Classification

Zero-Shot Text-to-Image Retrieval

使用 Usage

from PIL import Image
import requests
import clip
import torch
from Transformers import BertForSequenceClassification, BertConfig, BertTokenizer
from transformers import CLIPProcessor, CLIPModel
import numpy as np
query_texts = ["一只猫", "一只狗",'两只猫', '两只老虎','一只老虎']  # 这里是输入文本的，可以随意替换。
# 加载Taiyi 中文 text encoder
text_tokenizer = BertTokenizer.from_pretrained("IDEA-CCNL/Taiyi-CLIP-Roberta-large-326M-Chinese")
text_encoder = BertForSequenceClassification.from_pretrained("IDEA-CCNL/Taiyi-CLIP-Roberta-large-326M-Chinese").eval()
text = text_tokenizer(query_texts, return_tensors='pt', padding=True)['input_ids']
url = "http://images.cocodataset.org/val2017/000000039769.jpg"  # 这里可以换成任意图片的url
# 加载CLIP的image encoder
clip_model = CLIPModel.from_pretrained("openai/clip-vit-large-patch14")  
processor = CLIPProcessor.from_pretrained("openai/clip-vit-large-patch14")
image = processor(images=Image.open(requests.get(url, stream=True).raw), return_tensors="pt")
with torch.no_grad():
    image_features = clip_model.get_image_features(**image)
    text_features = text_encoder(text).logits
    # 归一化
    image_features = image_features / image_features.norm(dim=1, keepdim=True)
    text_features = text_features / text_features.norm(dim=1, keepdim=True)
    # 计算余弦相似度 logit_scale是尺度系数
    logit_scale = clip_model.logit_scale.exp()
    logits_per_image = logit_scale * image_features @ text_features.t()
    logits_per_text = logits_per_image.t()
    probs = logits_per_image.softmax(dim=-1).cpu().numpy()
    print(np.around(probs, 3))

引用 Citation

如果您在您的工作中使用了我们的模型，可以引用我们的论文：

If you are using the resource for your work, please cite the our paper:

@article{fengshenbang,
  author    = {Jiaxing Zhang and Ruyi Gan and Junjie Wang and Yuxiang Zhang and Lin Zhang and Ping Yang and Xinyu Gao and Ziwei Wu and Xiaoqun Dong and Junqing He and Jianheng Zhuo and Qi Yang and Yongfeng Huang and Xiayu Li and Yanghan Wu and Junyu Lu and Xinyu Zhu and Weifeng Chen and Ting Han and Kunhao Pan and Rui Wang and Hao Wang and Xiaojun Wu and Zhongshen Zeng and Chongpei Chen},
  title     = {Fengshenbang 1.0: Being the Foundation of Chinese Cognitive Intelligence},
  journal   = {CoRR},
  volume    = {abs/2209.02970},
  year      = {2022}
}

也可以引用我们的网站:

You can also cite our website:

@misc{Fengshenbang-LM,
  title={Fengshenbang-LM},
  author={IDEA-CCNL},
  year={2021},
  howpublished={\url{https://github.com/IDEA-CCNL/Fengshenbang-LM}},
}