This is a PyTorch implementation of the Trip-ROMA paper.
If you don't have python 3 environment:
conda create -n trip python=3.8
conda activate trip
Then install the required packages:
pip install -r requirements.txt
- Config
CONFIG
,DATA_DIR
andLOG_DIR
inscript/train.sh
orscript/train_ddp.sh
. We use DDP for ImageNet and STL10 by default. - Run
for single gpu training, or
sh scripts/train.sh
for DDP training.sh scripts/train_ddp.sh
- Config
CONFIG
,DATA_DIR
andLOG_DIR
inscript/eval.sh
. - Run
sh scripts/eval.sh
@article{
li2023triproma,
title={Trip-{ROMA}: Self-Supervised Learning with Triplets and Random Mappings},
author={Wenbin Li and Xuesong Yang and Meihao Kong and Lei Wang and Jing Huo and Yang Gao and Jiebo Luo},
journal={Transactions on Machine Learning Research},
issn={2835-8856},
year={2023}
}