MGrAC: Multi-Granularity Alignment-Aware Contrastive Learning for Open-World Semi-Supervised Learning

This repository is the official implementation of MGrAC.

Training

Below is an example demonstrating how to train a model on the CIFAR-100 dataset.

python train.py --dataset cifar100 --batch_size 256 --epochs 200 --base_lr 0.3

Acknowledgement

Our code are based on ORCA, TRSSL, TIDA. Thanks for their great work.