Sciscore

🚀 SciScore

SciScore是一个基于特定模型微调的工具，它以隐式提示和生成图像为输入，输出代表两者科学一致性的分数，有助于评估图像与科学描述的匹配程度。

🚀 快速开始

from transformers import AutoProcessor, AutoModel
from PIL import Image
import torch

device = "cuda"
processor_name_or_path = "Jialuo21/SciScore"
model_pretrained_name_or_path = "Jialuo21/SciScore"

processor = AutoProcessor.from_pretrained(processor_name_or_path)
model = AutoModel.from_pretrained(model_pretrained_name_or_path).eval().to(device)

def calc_probs(prompt, images):
    
    image_inputs = processor(
        images=images,
        padding=True,
        truncation=True,
        max_length=77,
        return_tensors="pt",
    ).to(device)
    
    text_inputs = processor(
        text=prompt,
        padding=True,
        truncation=True,
        max_length=77,
        return_tensors="pt",
    ).to(device)

    with torch.no_grad():
        image_embs = model.get_image_features(**image_inputs)
        image_embs = image_embs / torch.norm(image_embs, dim=-1, keepdim=True)
    
        text_embs = model.get_text_features(**text_inputs)
        text_embs = text_embs / torch.norm(text_embs, dim=-1, keepdim=True)
    
        scores = model.logit_scale.exp() * (text_embs @ image_embs.T)[0]
        probs = torch.softmax(scores, dim=-1)
    return probs.cpu().tolist()

pil_images = [Image.open("./examples/camera_1.png"), Image.open("./examples/camera_2.png")]
prompt = "A camera screen without electricity sits beside the window, realistic."
print(calc_probs(prompt, pil_images))

✨ 主要特性

SciScore在基础模型CLIP-H上进行微调，使用了Science-T2I数据集。它能够根据输入的隐式提示和生成图像，输出代表两者科学一致性的分数。

📚 详细文档

资源链接

• 项目网站
• arXiv论文
• GitHub代码库
• Huggingface: Science-T2I-S&C基准
• Huggingface: Science-T2I训练集

引用信息

@misc{li2025sciencet2iaddressingscientificillusions,
  title={Science-T2I: Addressing Scientific Illusions in Image Synthesis}, 
  author={Jialuo Li and Wenhao Chai and Xingyu Fu and Haiyang Xu and Saining Xie},
  year={2025},
  eprint={2504.13129},
  archivePrefix={arXiv},
  primaryClass={cs.CV},
  url={https://arxiv.org/abs/2504.13129}, 
}

📄 许可证

本项目采用Apache-2.0许可证。