Recently I code several projects with PyTorch, and find it really a light-weight and easy-to-use deep learning framework. After coding thousands of lines, a thought emerges in my mind about what we can do for improving code re-use and accelerating programming. That’s why I try to introduce torchsharp, a sharp knife for PyTorch.
Note that this package is still under early development, and I’ll add features continously in future. Issues ans Pull Requests are extremely welcomed.
torchsharp is a framework for PyTorch which provides a set of sharp utilities aiming at speeding up programming and encouraging code re-use. The repository consists of:
- torchsharp.data : Useful stuff about data operation such as dummy datasets and image tranforms for data argumentation, etc.
- torchsharp.model : Helpful fucntions about model training process such as initializer and metrics, etc.
- torchsharp.utils : Other tools like logger and timer, etc.
Apart from torchsharp, there’re also two auxiliary libraries for PyTorch -
torchzoo is a zoo of models and datasets for PyTorch. Most of them are used frequently in my research life but not provided by
This repository consists of:
- torchzoo.datasets : Data loaders for popular vision datasets.
- torchzoo.models : Definitions for popular model architectures.
pytorch-starter-kit is a demo project and quick starter kit for PyTorch.