PyTorch-Template
This project borrows ideas from Tensorflow-Project-Template, aimming to provide a easy and extensible template for pytorch projects.
Structure
├── datasets
(wrapper for datasets)
├── mains
(main.py for various datasets and tasks)
├── models
(network descriptions)
├── options
(template parser for command line inputs)
├── trainers
(define how model's forward / backward and logs)
├── utils
(toolbox for drawing and scheduling)
├── scripts
(scripts for replicating experiement results)
└── work
(default folder to store logs/models)
Usage examples
-
Train ResNet-20 (pre-activation) on CIFAR100
python main/cifar.py \ ~/torch_data/cifar \ --arch resnet20 \ --dataset cifar100 -
Train ResNet-18 on ImageNet
python main/imagenet.py \ ~/torch_data/imagenet \ --arch resnet18