Multiple teachers are beneficial lightweight and noise-resistant student models for point-of-care imaging classification

Introduction

Data Preparation

Download the dataset and store it in the File Organization

Run the files for data processing in the order written in the File Organization

Dataset

ISIC 2018 | BUSI | Dermnet

File Organization

├── code
    ├── dataloaders
        ├── BUSI_OverSampling      "BUSI second step"
        ├── BUSI_processing        "BUSI first step"
        ├── datasets               "pytorch dataloader"
        ├── Dermnet_OverSampling   "Dermnet second step"
        ├── Dermnet_processing     "Dermnet first step"
        ├── HairAugmentation       "ISIC2018 third step"
        ├── ISIC2018_OverSampling  "ISIC2018 second step"
        ├── ISIC2018_processing    "ISIC2018 first step"
        ├── merge_file_generate    "All Dataset merge to generate csv file"
        ├── resize                 "resize the hair images "
        └── utils                  "cal the std and mean of dataset,used for normalization"
    ├── networks                   
        ├── net_factory            "the factory of models"
        ├── ShiftMLP_base          "student model ---base"
        ├── ShiftMLP_small         "student model ---small"
        ├── SKAttention            "the SK Attention"
        ├── SwinTransformers       "the SwinTransformers"
        ├── UASwinTv2b             "the teacher model ---base"
        ├── UASwinTv2s             "the teacher model ---small"
        └── UASwinTv2t             "the teacher model ---tiny"
    ├── utils
        ├── losses
        ├── meterics
        ├── meterics2
        ├── meterics3
        ├── plots
        ├── ramps
        └── transforms
    ├── cal_parameter
    ├── test_BUSI                   "test the model in BUSI"
    ├── test_Dermnet                "test the model in Dermnet"
    ├── test_ISIC2018               "test the model in ISIC2018"
    ├── train_MTKD_BUSI             "train the model in BUSI"
    ├── train_MTKD_Dermnet          "train the model in Dermnet"
    ├── train_MTKD_ISIC2018         "train the model in ISIC2018"
    └── train_Teacher               "First train the Teacher model"
    
    ...
    "The UASwinT is the teacher model"
    "The name with KD means:use Knowledge distillation"
    "The name with MT means:use Mean Teacher model"
        
├── data [Your dataset path]
    ├── ISIC2018
        ├── ISIC2018_Task3_Test_GroundTruth
        ...
    ├── BUSI
        ├── benign
        ...
    ├── Dermnet
        ├── test
        ...
    ├── Hairs
        ...

Training and Testing

First Train the Teacher Model and Select The best model.

python -W ignore train_Teacher.py 

Then Train model with Global Teacher and Assistant Teacher

python -W ignore train_MTKD_ISIC2018.py 

Finally, Evaluation model

python -W ignore test_ISIC2018.py