noeltong / hotmap-classification

Hotmap Classification

This repository is a PyTorch implementation of hotmap classification.

Preliminaries

An Anaconda environment is needed. We first create a virtual env by

conda create -n cls python=3.9 --yes

and then activate it by

conda activate cls

After that, a torch==1.11.0 is suggested to be installed first.

conda install pytorch torchvision torchaudio cudatoolkit=11.3 -c pytorch --yes

and other requirements are listed in requirements.txt, which can be installed by a simple script

pip install -r requirements.txt

Usage of wandb

In this project, we used wandb to log information during training process. We install wandb by

pip install wandb

After the installation, we need to login to wandb using

wandb login

Training data

The training data is not available for public. However, you can still have your own data in .mat, stored in ./data folder. Don't forget to check ./utils/dataset.py to ensure there's no problem in loading the .mat file.

Logs and checkpoints

We also used tensorboard to record the loss and accuracy during training. The logs are stored in ./log.

We save checkppoints every 5 epochs in ./ckpt.