SIIM-ISIC Melanoma Classification: Identify melanoma in lesion images

Completed by HaoyiZhu and ShaofeiJiang

This is an competition on kaggle and is the big home work of the course: Practice of AI programming. The website of the competition is here

Requirements

• python 3.7
• pytorch 1.5.1
• torchvision 0.6.1
• pandas 1.0.5
• efficientnet_pytorch
• Windows

# You can install efficientnet_pytorch by:
pip install efficientnet_pytorch
# Or:
git clone https://github.com/lukemelas/EfficientNet-PyTorch
cd EfficientNet-Pytorch
pip install -e

Quick Start

• Get data:

Download the data from here and put them into ./data

• Training:

You can change the hyperparametrics such as lr, batch_size, begin_epoch, end_epoch, snapshop, etc. as you like in line 24 to line 30 in train.py.

Then simply run:

python train.py

• Testing:

If you want to use the model trained by yourself, just change the path in demo_inference.py.

Then simply run:

python demo_inference.py

Results Saving

• Training:

After every snapshot, a model will be saved in ./exp and simultaneously a result csv file will be saved in ./result

• Testing:

The result will be saved in ./result in the format of csv.