(IJCAI 2024) Bridging the Gap: Learning Pace Synchronization for Open-World Semi-Supervised Learning [Paper] [Website]
We are delighted that LPS has been accepted as a Long Oral presentation by IJCAI 2024.
This repository contains the implementation details of our Learning Pace Synchronization (LPS) approach for Open-World Semi-Supervised Learning
Bo Ye, Kai Gan, Tong Wei, Min-Ling Zhang, "Bridging the Gap: Learning Pace Synchronization for Open-World Semi-Supervised Learning"\
If you use the codes from this repo, please cite our work. Thanks!
@misc{ye2024bridging,
title={Bridging the Gap: Learning Pace Synchronization for Open-World Semi-Supervised Learning},
author={Bo Ye and Kai Gan and Tong Wei and Min-Ling Zhang},
year={2024},
eprint={2309.11930},
archivePrefix={arXiv},
primaryClass={cs.LG}
}
The code is built with following libraries:
For ImageNet 100, you need to utilize 'gen_imagenet_list.py' to generate the corresponding sample's list.
And the pretraining weights used in our paper can be downloaded in this link, which is provided by ORCA.
- To train on CIFAR-10, run
python lps_cifar.py --dataset cifar10 --labeled-num 5 --labeled-ratio 0.5- To train on CIFAR-100, run
python lps_cifar.py --dataset cifar100 --labeled-num 50 --labeled-ratio 0.5- To train on ImageNet-100, run
python lps_imagenet.py --dataset imagenet100 --labeled-num 50 --labeled-ratio 0.5