数据集获取

git clone https://github.com/arthursoenarto/csc490project.git

curl -L -o repo.zip https://github.com/arthursoenarto/csc490project/archive/refs/heads/main.zip
unzip repo.zip

源站 README 摘录（使用方式）

CSC490 Medical Imaging Dataset Project

Group Name: Team 8 - Skin Cancer Challengers

Team members:

• Arthur Alexandro Soenarto
• Gabriel El Haddad
• Xiaoning Wang
• Syed Taha Ali

Datasets:

• Skin Cancer MNIST HAM10000
• 2018 ISIC Challenge
  We choose the HAM10000 dataset, a large collection of multi-source dermatoscopic images of common pigmented skin lesions. It is also the dataset of the International Skin Imaging Collaboration (ISIC) 2018 challenge.

Set up Instructions

Run locally:

cd into folder then,

  $ virtualenv -p `which python3.8` venv/
  $ source venv/bin/activate
  $ pip install -r requirements.txt
  $ deactivate # when done

Run on Compute Canada vm:

  $ ssh host@graham.computecanada.ca
  $ virtualenv -p `which python3.8` venv/
  $ source venv/bin/activate
  $ pip install -r requirements.txt
  $ deactivate
  $ sbatch segtrainjob.sh # modify segtrainjob.sh with file u want to run
  $ squeue user=csc490w -t RUNNING # status of running job

If some of the libraries do not install correctly, need to download them using their Compute Canada alias

  $ avail_wheels "*name*"  # some libraries have different versions for cpu & gpu
  $ pip install <name> no-index

How to run/test/debug doubleunet/tripleunet locally:

Training (doubleunet_train.py):

under main() at end of file:

1. change the training images and ground truth file path based on directory structure
2. change batch_sizes and iter_sizes based on what you want to train on
3. change file path and name of where you want to store your pretrained models

at the end of def train():

1. under if plot:, change file path and name of where you want to store your loss curve

Testing (doubleunet_test.py for DoubleUNet, tripleunet_test.py for TripleUNet)

under main() at end of file:

1. change the training images and ground truth file path based on directory structure
2. change the validation images and ground truth file path based on directory structure
3. change doubleunet_models and model.load_state_dict() in the for loop to point to where the pretrained doubleunet models are based on directury structure
4. change unet_model_path to point to where the pretrained singleunet are based on directury struct

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/arthursoenarto/csc490project