Twin Contrastive Learning with Noisy Labels

This repo provides the official PyTorch implementation of our TCL accepted by CVPR 2023.

We have built new state-of-the-art performance on several benchmarked datasets.

Twin Contrastive Learning with Noisy Labels
https://arxiv.org/abs/2303.06930
Abstract: Learning from noisy data is a challenging task that significantly degenerates the model performance. In this paper, we present TCL, a novel twin contrastive learning model to learn robust representations and handle noisy labels for classification. Specifically, we construct a Gaussian mixture model (GMM) over the representations by injecting the supervised model predictions into GMM to link label-free latent variables in GMM with label-noisy annotations. Then, TCL detects the examples with wrong labels as the out-of-distribution examples by another two-component GMM, taking into account the data distribution. We further propose a cross-supervision with an entropy regularization loss that bootstraps the true targets from model predictions to handle the noisy labels. As a result, TCL can learn discriminative representations aligned with estimated labels through mixup and contrastive learning. Extensive experimental results on several standard benchmarks and real-world datasets demonstrate the superior performance of TCL. In particular, TCL achieves 7.5% improvements on CIFAR-10 with 90% noisy label -- an extremely noisy scenario.

If you found this code helps your work, do not hesitate to cite my paper or star this repo!

Introduction

The detail implementations and the results can be found in models/tcl, and in the following table 2, respectively.

EM Framework

Main Results

Training

Install requirements

git clone --depth 1 https://github.com/Hzzone/torch_clustering tmp && mv tmp/torch_clustering . && rm -rf tmp

pip install -r requirements.txt

Training Commands

The config files are in models/tcl/configs/, just run the following command:

export CUDA_VISIBLE_DEVICES=0,1,2,3 # use the first 4 GPUs
torchrun --master_port 17675 --nproc_per_node=4 main.py models/tcl/configs/cifar100_90_prer18.yml

We can also enable the WANDB to visualize the training!

Set the wandb parameters to true, and login to wandb.ai:

wandb login xxx

Download the pretrained models and training logs

The pretrained models can be saved by setting save_checkpoints to true.

Some training logs can be found in:

• Google Drive
• Baidu Disk

Citation

If you found this code or our work useful please cite us:

@inproceedings{huang2023twin,
  title={Twin Contrastive Learning with Noisy Labels},
  author={Huang, Zhizhong and Zhang, Junping and Shan, Hongming},
  booktitle={CVPR},
  year={2023}
}