rinna/日语CLIP-ViT-B-16模型

这是由rinna株式会社训练的日语CLIP（对比语言-图像预训练）模型。

其他可用模型请参见japanese-clip项目。

模型使用方法

1. 安装依赖包

$ pip install git+https://github.com/rinnakk/japanese-clip.git

2. 运行示例代码

import io
import requests
from PIL import Image
import torch
import japanese_clip as ja_clip

device = "cuda" if torch.cuda.is_available() else "cpu"

model, preprocess = ja_clip.load("rinna/japanese-clip-vit-b-16", cache_dir="/tmp/japanese_clip", device=device)
tokenizer = ja_clip.load_tokenizer()

img = Image.open(io.BytesIO(requests.get('https://images.pexels.com/photos/2253275/pexels-photo-2253275.jpeg?auto=compress&cs=tinysrgb&dpr=3&h=750&w=1260').content))
image = preprocess(img).unsqueeze(0).to(device)
encodings = ja_clip.tokenize(
    texts=["狗", "猫", "大象"],
    max_seq_len=77,
    device=device,
    tokenizer=tokenizer, # 可选参数。若不传入则每次加载tokenizer
)

with torch.no_grad():
    image_features = model.get_image_features(image)
    text_features = model.get_text_features(**encodings)
    
    text_probs = (100.0 * image_features @ text_features.T).softmax(dim=-1)

print("标签概率:", text_probs)  # 输出示例: [[1.0, 0.0, 0.0]]

模型架构

该模型采用ViT-B/16 Transformer架构作为图像编码器，使用12层BERT作为文本编码器。图像编码器初始化自AugReg vit-base-patch16-224模型。

训练数据

模型在CC12M数据集上进行训练，并将标题翻译为日语。

发布日期

2022年5月12日

引用方式

@misc{rinna-japanese-clip-vit-b-16,
    title = {rinna/日语CLIP-ViT-B-16模型},
    author = {Shing, Makoto and Zhao, Tianyu and Sawada, Kei},
    url = {https://huggingface.co/rinna/japanese-clip-vit-b-16}
}

@inproceedings{sawada2024release,
    title = {日语预训练模型发布},
    author = {Sawada, Kei and Zhao, Tianyu and Shing, Makoto and Mitsui, Kentaro and Kaga, Akio and Hono, Yukiya and Wakatsuki, Toshiaki and Mitsuda, Koh},
    booktitle = {2024年计算语言学与语言资源联合国际会议论文集(LREC-COLING 2024)},
    month = {5},
    year = {2024},
    pages = {13898--13905},
    url = {https://aclanthology.org/2024.lrec-main.1213},
    note = {\url{https://arxiv.org/abs/2404.01657}}
}

许可证

Apache 2.0许可证