Official Implementation of CVPR 2022 paper: "Mimicking the Oracle: An Initial Phase Decorrelation Approach for Class Incremental Learning"

Overview

(CVPR 2022) Mimicking the Oracle: An Initial Phase Decorrelation Approach for Class Incremental Learning ArXiv

This repo contains Official Implementation of our CVPR 2022 paper: Mimicking the Oracle: An Initial Phase Decorrelation Approach for Class Incremental Learning.

1. Abstract

Class Incremental Learning (CIL) aims at learning a classifier in a phase-by-phase manner, in which only data of a subset of the classes are provided at each phase. Previous works mainly focus on mitigating forgetting in phases after the initial one. However, we find that improving CIL at its initial phase is also a promising direction. Specifically, we experimentally show that directly encouraging CIL Learner at the initial phase to output similar representations as the model jointly trained on all classes can greatly boost the CIL performance. Motivated by this, we study the difference between a na"ively-trained initial-phase model and the oracle model. Specifically, since one major difference between these two models is the number of training classes, we investigate how such difference affects the model representations. We find that, with fewer training classes, the data representations of each class lie in a long and narrow region; with more training classes, the representations of each class scatter more uniformly. Inspired by this observation, we propose Class-wise Decorrelation (CwD) that effectively regularizes representations of each class to scatter more uniformly, thus mimicking the model jointly trained with all classes (i.e., the oracle model). Our CwD is simple to implement and easy to plug into existing methods. Extensive experiments on various benchmark datasets show that CwD consistently and significantly improves the performance of existing state-of-the-art methods by around 1% to 3%.


2. Instructions to Run Our Code

Current codebase only contain experiments on LUCIR with CIFAR100 and ImageNet100. Code reproducing results based on PODNet and AANet are based on their repo and will be coming soon!


CIFAR100 Experiments w/ LUCIR

No need to download the datasets, everything will be dealt with automatically.

For LUCIR baseline, simply first navigate under "src" folder and run:

bash exp_cifar_lucir.sh

For LUCIR + CwD, first navigate under "src" folder and run:

bash exp_cifar_lucir_cwd.sh

ImageNet100 Experiments w/ LUCIR

To run ImageNet100, please follow the following two steps:

Step 1:

download and extract imagenet dataset under "src/data/imagenet" folder.

Then, under "src/data/imagenet", run:

python3 gen_lst.py

This command will generate two list that determine the order of classes for class incremental learning. The class order is shuffled by seed 1993 like most previous works.


Step 2:

For LUCIR baseline, first navigate under "src" folder and run:

bash exp_im100_lucir.sh

For LUCIR+CWD, first navigate under "src" folder and run:

bash exp_im100_lucir_cwd.sh

Some Comments on Running Scripts.

For "SEED" variable in the scripts, it is not the seed that used to shuffle the class order, it is the seed that determines model initialisation/data loader sampling, etc. We vary "SEED" from 0,1,2 and average the Average Incremental Accuracy to obtain results reported in the paper.


3. For customized usage

To use our CwD loss in your own project, simply copy and paste the CwD loss implemented in "src/approach/aux_loss.py" will be fine.


4. Citation

If you find our repo/paper helpful, please consider citing our work :)

@article{shi2021mimicking,
  title={Mimicking the Oracle: An Initial Phase Decorrelation Approach for Class Incremental Learning},
  author={Shi, Yujun and Zhou, Kuangqi and Liang, Jian and Jiang, Zihang and Feng, Jiashi and Torr, Philip and Bai, Song and Tan, Vincent YF},
  journal={arXiv preprint arXiv:2112.04731},
  year={2021}
}

5. Contact

Yujun Shi ([email protected])

6. Acknowledgements

Our code is based on FACIL, one of the most well-written CIL library in my opinion:)

7. Some Additional Remarks

Based on the original implementation of FACIL, I also implemented Distributed Data Parallel to enable multi-GPU training. However, it seems that the performance is not as good as single card training (about 0.5% lower). Therefore, in all experiments, I still use single card training.

Owner
Yujun Shi
PhD @ NUS
Yujun Shi
VolumeGAN - 3D-aware Image Synthesis via Learning Structural and Textural Representations

VolumeGAN - 3D-aware Image Synthesis via Learning Structural and Textural Representations 3D-aware Image Synthesis via Learning Structural and Textura

GenForce: May Generative Force Be with You 116 Dec 26, 2022
Code for Greedy Gradient Ensemble for Visual Question Answering (ICCV 2021, Oral)

Greedy Gradient Ensemble for De-biased VQA Code release for "Greedy Gradient Ensemble for Robust Visual Question Answering" (ICCV 2021, Oral). GGE can

21 Jun 29, 2022
ST++: Make Self-training Work Better for Semi-supervised Semantic Segmentation

ST++ This is the official PyTorch implementation of our paper: ST++: Make Self-training Work Better for Semi-supervised Semantic Segmentation. Lihe Ya

Lihe Yang 147 Jan 03, 2023
Learned image compression

Overview Pytorch code of our recent work A Unified End-to-End Framework for Efficient Deep Image Compression. We first release the code for Variationa

Jiaheng Liu 163 Dec 04, 2022
NeoDTI: Neural integration of neighbor information from a heterogeneous network for discovering new drug-target interactions

NeoDTI NeoDTI: Neural integration of neighbor information from a heterogeneous network for discovering new drug-target interactions (Bioinformatics).

62 Nov 26, 2022
[CVPR 2021] MetaSAug: Meta Semantic Augmentation for Long-Tailed Visual Recognition

MetaSAug: Meta Semantic Augmentation for Long-Tailed Visual Recognition (CVPR 2021) arXiv Prerequisite PyTorch = 1.2.0 Python3 torchvision PIL argpar

51 Nov 11, 2022
State-to-Distribution (STD) Model

State-to-Distribution (STD) Model In this repository we provide exemplary code on how to construct and evaluate a state-to-distribution (STD) model fo

<a href=[email protected]"> 2 Apr 07, 2022
Log4j JNDI inj. vuln scanner

Log-4-JAM - Log 4 Just Another Mess Log4j JNDI inj. vuln scanner Requirements pip3 install requests_toolbelt Usage # make sure target list has http/ht

Ashish Kunwar 66 Nov 09, 2022
Deep Learning to Create StepMania SM FIles

StepCOVNet Running Audio to SM File Generator Currently only produces .txt files. Use SMDataTools to convert .txt to .sm python stepmania_note_generat

Chimezie Iwuanyanwu 8 Jan 08, 2023
A Pytorch Implementation of [Source data‐free domain adaptation of object detector through domain

A Pytorch Implementation of Source data‐free domain adaptation of object detector through domain‐specific perturbation Please follow Faster R-CNN and

1 Dec 25, 2021
Kaggle G2Net Gravitational Wave Detection : 2nd place solution

Kaggle G2Net Gravitational Wave Detection : 2nd place solution

Hiroshechka Y 33 Dec 26, 2022
PyTorch 1.0 inference in C++ on Windows10 platforms

Serving PyTorch Models in C++ on Windows10 platforms How to use Prepare Data examples/data/train/ - 0 - 1 . . . - n examples/data/test/

Henson 88 Oct 15, 2022
Provide baselines and evaluation metrics of the task: traffic flow prediction

Note: This repo is adpoted from https://github.com/UNIMIBInside/Smart-Mobility-Prediction. Due to technical reasons, I did not fork their code. Introd

Zhangzhi Peng 11 Nov 02, 2022
PIGLeT: Language Grounding Through Neuro-Symbolic Interaction in a 3D World [ACL 2021]

piglet PIGLeT: Language Grounding Through Neuro-Symbolic Interaction in a 3D World [ACL 2021] This repo contains code and data for PIGLeT. If you like

Rowan Zellers 51 Oct 08, 2022
Implementation of paper "Self-supervised Learning on Graphs:Deep Insights and New Directions"

SelfTask-GNN A PyTorch implementation of "Self-supervised Learning on Graphs: Deep Insights and New Directions". [paper] In this paper, we first deepe

Wei Jin 85 Oct 13, 2022
Code for the Lovász-Softmax loss (CVPR 2018)

The Lovász-Softmax loss: A tractable surrogate for the optimization of the intersection-over-union measure in neural networks Maxim Berman, Amal Ranne

Maxim Berman 1.3k Jan 04, 2023
Fast and customizable reconnaissance workflow tool based on simple YAML based DSL.

Fast and customizable reconnaissance workflow tool based on simple YAML based DSL, with support of notifications and distributed workload of that work

Américo Júnior 3 Mar 11, 2022
Sequence lineage information extracted from RKI sequence data repo

Pango lineage information for German SARS-CoV-2 sequences This repository contains a join of the metadata and pango lineage tables of all German SARS-

Cornelius Roemer 24 Oct 26, 2022
DeconvNet : Learning Deconvolution Network for Semantic Segmentation

DeconvNet: Learning Deconvolution Network for Semantic Segmentation Created by Hyeonwoo Noh, Seunghoon Hong and Bohyung Han at POSTECH Acknowledgement

Hyeonwoo Noh 325 Oct 20, 2022