M3D CLIP

🚀 M3D-CLIP

M3D-CLIP is a 3D medical CLIP model that aligns vision and language, offering powerful features for 3D medical image and text retrieval tasks.

✨ Features

• M3D-CLIP is part of the M3D series.
• It aligns vision and language via contrastive loss on the M3D-Cap dataset.
• The vision encoder uses 3D ViT with an image size of 32256256 and a patch size of 41616.
• The language encoder initializes with a pre - trained BERT.
• It can be used for 3D medical image and text retrieval tasks.
• It provides aligned and powerful image and text features for downstream tasks.
• Text - aligned visual encoders serve as excellent pre - trained models for visual and multi - modal tasks.

🚀 Quick Start

device = torch.device("cuda") # or cpu

tokenizer = AutoTokenizer.from_pretrained(
    "GoodBaiBai88/M3D-CLIP",
    model_max_length=512,
    padding_side="right",
    use_fast=False
)
model = AutoModel.from_pretrained(
    "GoodBaiBai88/M3D-CLIP",
    trust_remote_code=True
)
model = model.to(device=device)

# Prepare your 3D medical image:
# 1. The image shape needs to be processed as 1*32*256*256, considering resize and other methods.
# 2. The image needs to be normalized to 0-1, considering Min-Max Normalization.
# 3. The image format needs to be converted to .npy 
# 4. Although we did not train on 2D images, in theory, the 2D image can be interpolated to the shape of 1*32*256*256 for input.
    
image_path = ""
input_txt = ""

text_tensor = tokenizer(input_txt, max_length=512, truncation=True, padding="max_length", return_tensors="pt")
input_id = text_tensor["input_ids"].to(device=device)
attention_mask = text_tensor["attention_mask"].to(device=device)
image = np.load(image_path).to(device=device)

with torch.inference_mode():
    image_features = model.encode_image(image)[:, 0]
    text_features = model.encode_text(input_id, attention_mask)[:, 0]

📄 License

This project is licensed under the Apache - 2.0 license.

📚 Documentation

Citation

If you find our work helpful, please consider citing the following work:

@misc{bai2024m3d,
      title={M3D: Advancing 3D Medical Image Analysis with Multi-Modal Large Language Models}, 
      author={Fan Bai and Yuxin Du and Tiejun Huang and Max Q. -H. Meng and Bo Zhao},
      year={2024},
      eprint={2404.00578},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}

📦 Information Table