数据集获取

git clone https://github.com/ku262/EndoFinder-Scene.git

curl -L -o repo.zip https://github.com/ku262/EndoFinder-Scene/archive/refs/heads/main.zip
unzip repo.zip

源站 README 摘录（使用方式）

EndoFinder: Online Lesion Retrieval for Explainable Colorectal Polyp Diagnosis Leveraging Latent Scene Representations

A preliminary version of this work was published at MICCAI 2024. Code is available at here.
This work significantly extends that work by moving from 2D image retrieval to a 2.5D/3D scene retrieval paradigm. We achieve this by explicitly modeling polyps as scenes and leveraging multi-view
perspectives to capture implicit 3D information within a latent
scene representation.
<div align=“center”>

---

Setup

• Environment

  An example of installation commands is provided as follows:

  git clone git@github.com:ku262/EndoFinder-Scene.git
  cd EndoFinder-Scene
  pip3 install torch torchvision torchaudio 
  pip3 install -r requirements.txt
  

• Model Preparation

  We will release the trained models from our original experiments to
  facilitate reproducibility. The release of EndoFinder-Scene.pth is planned upon acceptance of the paper.

• Data Preparation

  We introduce PolypScene-250, a novel multi-view polyp dataset, including the PolypScene-80 subset with detailed histopathology annotations.

---

Sampling

• Train

  python train.py \
            batch_size=512 \
            epochs=200 \
            output_dir=outputs \
            finetune=EndoFinder.pth \
  

• Val

  python demo.py \
            batch_size=64 \
            resume=EndoFinder_Scene.pth \
  

数据加载示例（图像类）

from PIL import Image
import glob, os

files = (glob.glob(os.path.join(path, "**", "*.png"), recursive=True)
       + glob.glob(os.path.join(path, "**", "*.jpg"), recursive=True)
       + glob.glob(os.path.join(path, "**", "*.tif"), recursive=True))
print("图像文件数:", len(files))
img = Image.open(files[0]); print("尺寸/模式:", img.size, img.mode)

# torchvision Dataset 方式:
# from torchvision import datasets
# ds = datasets.ImageFolder(path)  # 要求 子目录=类别

目录组织与标注格式以源站说明和下载后实际文件为准。

完整仓库：github.com/ku262/EndoFinder-Scene