rohinarora/minSSL

Minimal PyTorch Re-Implementation of Semi Supervised Learning Algorithms

minSSL - A Minimal Library for Self-Supervised Learning

Installation

• Create pytorch env as mentioned in https://rc-docs.northeastern.edu/en/latest/using-discovery/workingwithgpu.html#using-gpus-with-pytorch

$ conda activate <env_name>

Usage

1. Pretrain

python pre_train.py --help #run options
python pre_train.py #default config 
python pre_train.py -data ./datasets --dataset-name stl10 --log-every-n-steps 100 --epochs 100 

2. Evaluate on downstream tasks

python downstream_eval.py --help #run options

• Currently supports SimCLR and MoCo. More SSL algorithms coming soon !

Authors

• Rohin Arora, Syed Shahbaaz Ahmed and Varun Sahasrabudhe