Sample Prior Guided Robust Model Learning to Suppress Noisy Labels

Related tags

Deep Learning PGDF

Overview

PGDF

This repo is the official implementation of our paper "Sample Prior Guided Robust Model Learning to Suppress Noisy Labels ".

Citation

If you use this code/data for your research, please cite our paper "Sample Prior Guided Robust Model Learning to Suppress Noisy Labels ".

@misc{chen2021sample,
      title={Sample Prior Guided Robust Model Learning to Suppress Noisy Labels}, 
      author={Wenkai Chen and Chuang Zhu and Yi Chen},
      year={2021},
      eprint={2112.01197},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}

Training

Take CIFAR-10 with 50% symmetric noise as an example:

First, please modify the data_path in presets.json to indicate the location of your dataset.

Then, run

python train_cifar_getPrior.py --preset c10.50sym

to get the prior knowledge. Related files will be saved in checkpoints/c10/50sym/saved/.

Next, run

python train_cifar.py --preset c10.50sym

for the subsequent training process.

c10 means CIFAR-10, 50sym means 50% symmetric noise.
Similarly, if you want to take experiment on CIFAR-100 with 20% symmetric noise, you can use the command:

python train_cifar_getPrior.py --preset c100.20sym

python train_cifar.py --preset c100.20sym

Additional Info

The (basic) semi-supervised learning part of our code is borrow from the official DM-AugDesc implementation.

Since this paper has not yet been submitted, we only release part of the experimental code. We will release all the experimental code after this paper is accepted by a conference.

Comments

Requirements

An installation file with the requirements would be handy to install all the dependencies with pip.

I believe that there is a simple command to generate a requirements.txt given an environment.

Thanks!

opened by aldakata 3
details about mini webvision1.0 dataset

Thanks for your brilliant work first. Actually we try to replicate your results, so i was wondering that how many samples does the mini Webvision 1.0 dataset your used in this work contain? Is that 61K? I was so confused because some works state that they also use a subset of the first 50 classes in webvision1.0 which contains more than 100K samples in training set, such as Aum[https://arxiv.org/abs/2001.10528]. Looking forward for your kindly help

opened by Christophe-Jia 2
GMM after epoch 200

After reading the paper:

Specifically, as shown in Figure 2, at each epoch, we get the clean probability wit of each sample from training loss by using Gaussian Mixture Model (GMM)

But on the code this GMM is only accounted for after epoch 200. Is this an error or am I missing something? https://github.com/bupt-ai-cz/PGDF/blob/f7b6de71c0959c8a268a74da126a4e1858aeb290/train_cifar.py#L471

Thanks in advance!

opened by aldakata 1

PyTorch implemention of ICCV'21 paper SGPA: Structure-Guided Prior Adaptation for Category-Level 6D Object Pose Estimation

SGPA: Structure-Guided Prior Adaptation for Category-Level 6D Object Pose Estimation This is the PyTorch implemention of ICCV'21 paper SGPA: Structure

24 Dec 5, 2022

Simple and Robust Loss Design for Multi-Label Learning with Missing Labels

Simple and Robust Loss Design for Multi-Label Learning with Missing Labels Official PyTorch Implementation of the paper Simple and Robust Loss Design

28 Oct 27, 2022

Yolox-bytetrack-sample - Python sample of MOT (Multiple Object Tracking) using YOLOX and ByteTrack

yolox-bytetrack-sample YOLOXとByteTrackを用いたMOT(Multiple Object Tracking)のPythonサン

12 Nov 9, 2022

Code for 'Self-Guided and Cross-Guided Learning for Few-shot segmentation. (CVPR' 2021)'

SCL Introduction Code for 'Self-Guided and Cross-Guided Learning for Few-shot segmentation. (CVPR' 2021)' We evaluated our approach using two baseline

34 Oct 8, 2022

Training a deep learning model on the noisy CIFAR dataset

Training-a-deep-learning-model-on-the-noisy-CIFAR-dataset This repository contai

1 Jun 14, 2022

An original implementation of "Noisy Channel Language Model Prompting for Few-Shot Text Classification"

Channel LM Prompting (and beyond) This includes an original implementation of Sewon Min, Mike Lewis, Hannaneh Hajishirzi, Luke Zettlemoyer. "Noisy Cha

92 Jan 7, 2023

Towards Calibrated Model for Long-Tailed Visual Recognition from Prior Perspective

Towards Calibrated Model for Long-Tailed Visual Recognition from Prior Perspective Zhengzhuo Xu, Zenghao Chai, Chun Yuan This is the PyTorch implement

16 Dec 15, 2022

Super Pix Adv - Offical implemention of Robust Superpixel-Guided Attentional Adversarial Attack (CVPR2020)

Super_Pix_Adv Offical implemention of Robust Superpixel-Guided Attentional Adver

8 Oct 26, 2022

This YoloV5 based model is fit to detect people and different types of land vehicles, and displaying their density on a fitted map, according to their coordinates and detected labels.

This YoloV5 based model is fit to detect people and different types of land vehicles, and displaying their density on a fitted map, according to their

8 May 22, 2022

Sample Prior Guided Robust Model Learning to Suppress Noisy Labels

Related tags

Overview

PGDF

Citation

Training

Additional Info

You might also like...

PyTorch implemention of ICCV'21 paper SGPA: Structure-Guided Prior Adaptation for Category-Level 6D Object Pose Estimation

Simple and Robust Loss Design for Multi-Label Learning with Missing Labels

Yolox-bytetrack-sample - Python sample of MOT (Multiple Object Tracking) using YOLOX and ByteTrack

Code for 'Self-Guided and Cross-Guided Learning for Few-shot segmentation. (CVPR' 2021)'

Training a deep learning model on the noisy CIFAR dataset

An original implementation of "Noisy Channel Language Model Prompting for Few-Shot Text Classification"

Towards Calibrated Model for Long-Tailed Visual Recognition from Prior Perspective

Super Pix Adv - Offical implemention of Robust Superpixel-Guided Attentional Adversarial Attack (CVPR2020)

This YoloV5 based model is fit to detect people and different types of land vehicles, and displaying their density on a fitted map, according to their coordinates and detected labels.

Comments

Requirements

details about mini webvision1.0 dataset

GMM after epoch 200

Owner

CVSM Group - email: [email protected]

Official PyTorch implemention of our paper "Learning to Rectify for Robust Learning with Noisy Labels".

DIT is a DTLS MitM proxy implemented in Python 3. It can intercept, manipulate and suppress datagrams between two DTLS endpoints and supports psk-based and certificate-based authentication schemes (RSA + ECC).

PyTorch implementation of "Contrast to Divide: self-supervised pre-training for learning with noisy labels"

A curated (most recent) list of resources for Learning with Noisy Labels

NeurIPS 2021, "Fine Samples for Learning with Noisy Labels"

Generalized Jensen-Shannon Divergence Loss for Learning with Noisy Labels

A GOOD REPRESENTATION DETECTS NOISY LABELS

A Light CNN for Deep Face Representation with Noisy Labels

Large-Scale Pre-training for Person Re-identification with Noisy Labels (LUPerson-NL)

PFENet: Prior Guided Feature Enrichment Network for Few-shot Segmentation (TPAMI).