Robust Contrastive Learning against Noisy Views
This repository provides a PyTorch implementation of the Robust InfoNCE loss proposed in paper Robust Contrastive Learning against Noisy Views.
Requirements:
- PyTorch >= 1.5.0
Pseudo Code
The implementation only requires a small modification to the InfoNCE code.
# bsz : batch size (number of positive pairs)
# pos : exponent for positive example, shape=[bsz]
# neg : sum of exponents for negative examples, shape=[bsz]
# q, lam : hyperparameters of RINCE
info_nce_loss = -log(pos / (pos + neg))
rince_loss = -pos**q / q + (lam * (pos + neg))**q / q
ImagNet Experiments
The code can be found in ImageNet/moco-v3 and ImageNet/simclr.
Citation
If you find this repo useful for your research, please consider citing the paper
@article{chuang2022robust,
title={Robust Contrastive Learning against Noisy Views},
author={Chuang, Ching-Yao and Hjelm, R Devon and Wang, Xin and Vineet, Vibhav and Joshi, Neel and Torralba, Antonio and Jegelka, Stefanie and Song, Yale},
journal={arXiv preprint arXiv:2201.04309},
year={2022}
}
For any questions, please contact Ching-Yao Chuang ([email protected]).
57 Dec 4, 2022
469 Dec 26, 2022
321 Jan 9, 2023
27 Dec 23, 2022
13 Dec 22, 2022
2 Dec 9, 2021
1 Jun 14, 2022
1 Oct 26, 2021
![The official implementation of NeMo: Neural Mesh Models of Contrastive Features for Robust 3D Pose Estimation [ICLR-2021]. https://arxiv.org/pdf/2101.12378.pdf](https://github.com/Angtian/NeMo/raw/main/example.gif)
76 Nov 23, 2022
22 Dec 23, 2022
202 Dec 13, 2022
3 Nov 19, 2022
31 Dec 1, 2022
33 Nov 1, 2022
28 Dec 12, 2022
111 Dec 29, 2022
44 Nov 20, 2022
35 Dec 26, 2022
55 Nov 23, 2022