/SimpleCV

Simplify training, evaluation, prediction in Pytorch

Primary LanguagePythonMIT LicenseMIT

DOI




This repo is aimed to simplify training, evaluation and prediction in Pytorch.

Features

  1. Focus on your research rather than training template codes
  2. Dynamic module registration mechanism makes you customize components on the fly
  3. Flexible plugin mechanism for a hackable trainer without any coupling!
  4. High performance parallel training using Pytorch
  5. Support mixed precision training, significantly reducing GPU memory usage with similar performance
  6. Support stable distribute training and Sync BN by offical repo and NVIDIA/apex

Installation

pip install --upgrade git+https://github.com/Z-Zheng/SimpleCV.git

Requirements:

  • pytorch >= 1.1.0
  • tensorboardX
  • opencv, skimage, sklearn, pillow

Citing SimpleCV

If you use SimpleCV in your research, please use the following BibTeX entry.

@misc{simplecv2018,
  author =       {Zhuo Zheng},
  title =        {SimpleCV},
  howpublished = {\url{https://github.com/Z-Zheng/SimpleCV}},
  year =         {2018}
}

Usage

Please refer to USAGE.md for the basic usage of SimpleCV.

Projects using SimpleCV

Change logs

  • 2019/10/29 v0.3.4 released! More preset models have been added.
  • 2019/06/25 v0.3.1 released! More features have been added.
  • 2019/05/24 v0.3.0 released!
  • 2019/05/05 compatible with pytorch==1.1.0 (naive sync bn in ddp train)
  • 2019/04/08 v0.2.0 released! Support apex!
  • 2019/01/25 summary grads and weights
  • 2018/12/20 support SE Block