Steering Self-Supervised Feature Learning Beyond Local Pixel Statistics [Project Page]

Simon Jenni, Hailin Jin, and Paolo Favaro.
In CVPR, 2020.

This repository contains code for self-supervised pre-training and supervised transfer learning on the STL-10 dataset.

Training and evaluation on ImageNet is coming soon!

Requirements

The code is based on Python 3.7 and tensorflow 1.15.

How to use it

1. Setup

• Set the paths to the data and log directories in constants.py.
• Run init_datasets.py to download and convert the STL-10 dataset to the TFRecord format:

python init_datasets.py

2. Training and evaluation

• To train and evaluate a transformation classifier on STL-10 execute run_stl10.py. An example usage could look like this:

python run_stl10.py --tag='test' --num_gpus=1

Citation

If you find this repository useful for your research, please use the following.

@inproceedings{jenni2020steering,
  title={Steering Self-Supervised Feature Learning Beyond Local Pixel Statistics},
  author={Jenni, Simon and Jin, Hailin and Favaro, Paolo},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  pages={6408--6417},
  year={2020}
}