template.md

If you want to use the pytorch project template

1. Define your own dataloader

Define your own dataloader, please inherit BaseDataLoader in ./base/base_data_loader.py (instead of the official version torch.utils.data.DataLoader), this will save you much time and workload.

Having trouble with how to define ? Read ./data_loader/data_loaders.pymay give you some hint!

2. Define your loss and metric

For classification, using cls_loss in ./models/loss.py and accuracy in ./models/metric.py is just fine.

3. Define your model

There are a lot of models you can use dircetly, or you can define it youself.

4. Train it

It's recommended to use the classifier in classifier.py, it's really convenient to train, resume , and very flexible to change default settings (learning rate cheduler, optimizer, epoch...)