Semantic Segmentation of Pathological Images

中文

Implemented by Pytorch, it can also be a tutorial of semantic segmentation.

If you like it, please star me. 😄

Datasets

There are some datasets from kaggle and other websites.

• Brain MRI Segmentation
• Chest Xray Masks and Labels
• Skin Lesion Segmantation
• CHAOS
• Glomerulus

Models

• UNet
• UNet++
• Attention UNet
• R2U-Net
• Attention R2U-Net
• NFNPlus
• XNet

Requirements

• Pytorch 1.4.0
• Torchvision 0.5.0
• Python 3.7
• Some other libraries (find what you miss when running the code :-P)

Preparations

• After downloading datasets, generate the train.json and val.json for each dataset with the following format:

[
    {
        "image_path": "/home/tangwenqi/workspace/data/small_g4/coco_train/JPEGImages/20180550-PASM.jpg",
        "mask_path": "/home/tangwenqi/workspace/data/small_g4/coco_train/Labels/20180550-PASM.png"
    },
    {
        "image_path": "/home/tangwenqi/workspace/data/small_g4/coco_train/JPEGImages/20180620-PASM.jpg",
        "mask_path": "/home/tangwenqi/workspace/data/small_g4/coco_train/Labels/20180620-PASM.png"
    },
]

• For each dataset, modify its absolute path in corresponding class.

Usages

run train.py with some required and optional parameters, for more details please look train.py.