数据集获取

git clone https://github.com/catai9/Cancer-Net-SCa-Synth.git

curl -L -o repo.zip https://github.com/catai9/Cancer-Net-SCa-Synth/archive/refs/heads/main.zip
unzip repo.zip

源站 README 摘录（使用方式）

Cancer-Net-SCa-Synth

This is the repo containing scripts and utilities for generating images of benign and malignant skin lesions using the Stable Diffusion model with DreamBooth. The latest version producing the best results uses scripts from HuggingFace which are licensed under the Apache 2.0 license (as of 2024-05-26).

Project structure

.
├── data # Training data, recommend symlinking
│   ├── jpeg # JPEG images
│   │   ├── test # Test set
│   │   └── train # Training set
│   ├── test.csv # Metadata CSV for test set
│   └── train.csv # Metadata CSV for training set
├── src # Python source code
│   └── create_training_dataset.py # Script to generate folder of images for training SD using DreamBooth trainer
│   └── train_dreambooth.py # HuggingFace Python script for training SD using DreamBooth trainer
│   └── generate_images.py # Generate images using the trained SD with DreamBooth trainer
│   └── create_generated_csv_file.py # Create a csv file for the generated images
│   └── prepreprocess_data.py # Preprocess data for model training
│   └── train_mobilenetv2_model.py # Train MobileNetV2 model
# Misc individual files
├── requirements.txt
└── README.md

Setup

The following setup should be done on a machine with a Nvidia GPU.

1. Download/prepare the ISIC 2020 skin lesion dataset
   The dataset can be downloaded here.
   To download from Kaggle:

kaggle competitions download -c siim-isic-melanoma-classification # Need Kaggle CLI installed

then unzip the file and create a symlink to the folder under the name data/ in the project root:

ln -s <downloaded folder> data

2. Create a directory containing the desired training instances. A helper script can be used to generate this directory, sampling random benign/malignant images as follows:

python src/create_training_dataset.py \
    malignant \ 
    source_csv data/train.csv \
    source_dir data/jpeg/train \
    target_dir data/generated/train-malignant \
    subset_size 300

python src/create_training_dataset.py \
    benign \ 
    source_csv data/train.csv \
    source_dir data/jpeg/train \
    target_dir data/generated/train-benign \
    subset_size 300

Please run python src/create_training_dataset.py -h for more information on the script usage and available flags.

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/catai9/Cancer-Net-SCa-Synth