SlideShare a Scribd company logo
The Majority Can Help the Minority:
Context-rich Minority Oversampling for Long-tailed Classification
1Seoul National University, 2NAVER AI Lab
Poster ID: 159a & Poster Time: 22, Jun. 10:00-12:30
Seulki Park1 Byeongho Heo2 Sangdoo Yun2 Jin Young Choi1
Youngkyu Hong2
Long-tailed Classification
2
Introduction Proposed Method Experiment Conclusion
Many real-world data often exhibit long-tailed distribution.
✓ The model trained on such imbalanced data tends to overfit the majority classes.
✓ That is, the model performs poorly on minority classes.
Problem Definition:
● Input: Long-tailed (imbalanced) training data & uniform-distributed (balanced) test data.
● Goal: To make a robust model that can generalize well on balanced test data.
The Majority Can Help the Minority: Context-rich Minority Oversampling for Long-tailed Classification
Faces (Zhang et al., 2017) Places (Wang et al., 2017) Species (Van Horn et al., 2018) Actions (Zhang et al., 2019)
* Images by authors.
Previous Oversampling Methods
Introduction Proposed Method Experiment Conclusion
1. Random Oversampling (ROS)
◦ A simple and straightforward method which repeatedly oversample minor classes.
◦ However, this may intensify overfitting problem [1].
2. Synthetic Minority Over-sampling Technique (SMOTE), 2002
◦ Oversamples minority samples by interpolating between existing minority samples and their nearest minority neighbors.
◦ However, difficulties for end-to-end algorithm and largescale image datasets due to the high computational complexity of
calculating K-Nearest Neighbor for every sample.
3
The Majority Can Help the Minority: Context-rich Minority Oversampling for Long-tailed Classification
[1] Deep imbalanced attribute classification using visual attention aggregation, ECCV, 2018.
SMOTE Figure from: http://www.incodom.kr/SMOTE
Previous Oversampling Methods
Introduction Proposed Method Experiment Conclusion
3. Generative Adversarial Minority Oversampling (GAMO), 2019
◦ Produces new minority samples by training a convex generator, inspired by the success of
generative adversarial networks (GANs).
◦ However, difficult to train a generator (mode collapse) & additional training cost.
4. MetaSAug: Meta Semantic Augmentation for Long-Tailed Visual Recognition, 2021
◦ Uses implicit semantic data augmentation (ISDA) algorithm [1].
◦ However, this meta-learning-based method requires additional balanced validation, and hundreds and thousands of
iterations for training (high training cost).
4
The Majority Can Help the Minority: Context-rich Minority Oversampling for Long-tailed Classification
[1] Implicit semantic data augmentation for deep networks, NeurIPS, 2019.
Previous Oversampling Methods
Introduction Proposed Method Experiment Conclusion
Random Oversampling
◦ A simple and straightforward method which repeatedly oversample minor classes.
◦ However, this may intensify overfitting problem [1].
Synthetic Minority Over-sampling Technique (SMOTE), 2002
◦ Oversamples minority samples by interpolating between existing minority samples and their nearest minority neighbors.
◦ However, difficulties for end-to-end algorithm and largescale image datasets due to the high computational complexity of
calculating K-Nearest Neighbor for every sample.
Generative Adversarial Minority Oversampling (GAMO), 2019
◦ Produces new minority samples by training a convex generator, inspired by the success of generative adversarial networks
(GANs).
◦ However, difficult to train a generator (mode collapse) & additional training cost.
MetaSAug: Meta Semantic Augmentation for Long-Tailed Visual Recognition, 2021
◦ Uses implicit semantic data augmentation (ISDA) algorithm [1].
◦ However, this meta-learning-based method requires additional balanced validation, and hundreds and thousands of
iterations for training (high training cost).
5
The Majority Can Help the Minority: Context-rich Minority Oversampling for Long-tailed Classification
Limitation of previous methods:
1) Simple methods generate only context-limited images. (e.g, ROS, SMOTE)
- Limited improvement especially when imbalance is severe.
2) Recent methods require additional expensive training cost (e.g, GAMO, MetaSAug)
- E.g., training generator, additional balanced validation set, longer training epochs.
→ We need ‘Simple & Context-rich’ oversampling method!
Motivation
Introduction Proposed Method Experiment Conclusion
Q. How can we generate diverse ‘context-rich minority samples’ from long-tailed distribution?
A. Let’s pay attention to the characteristics of long-tailed distributions.
Key Observations:
✓ Majority class samples are data-rich and information-rich!
→ Let’s use the affluent information of the majority samples
to generate new minority samples.
Key Idea
- We can use the rich major-class images as the background
for the newly created minor-class images.
6
The Majority Can Help the Minority: Context-rich Minority Oversampling for Long-tailed Classification
Proposed Method: Context-rich Minority Oversampling (CMO)
Introduction Proposed Method Experiment Conclusion
Recap: CutMix (Yun et al., 2019)
- A simple but effective data augmentation method used in many visual tasks.
෤
𝑥 = 𝑴⨀𝑥𝑏
+ 𝟏 − 𝑴 ⨀𝑥𝑓
, ෤
𝑦 = 𝜆𝑦𝑏
+ 1 − 𝜆 𝑦𝑓
𝑥𝑏
, 𝑦𝑏
, 𝑥𝑓
, 𝑦𝑓
~ 𝑃
𝑴 ∈ 0, 1 𝑊×𝐻
: a binary mask
→ designed for a class balanced dataset.
Naively using CutMix generates more samples of the majority classes.
Context-rich Minority Oversampling (CMO)
- For an imbalanced dataset, we use different distributions for background and foreground images.
𝑥𝑏
, 𝑦𝑏
~ 𝑃, 𝑥𝑓
, 𝑦𝑓
~ 𝑄
𝑄 : minor-class-weighted distribution.
7
The Majority Can Help the Minority: Context-rich Minority Oversampling for Long-tailed Classification
CutMix
Comparison with CutMix [3]
Proposed Method: Minor-class-weighted distribution
Introduction Proposed Method Experiment Conclusion
How to design minor-class-weighted sampling strategies?
- Re-weighting methods have provided a way how to assign appropriate weights to samples.
- Commonly used sampling strategies give a weight inversely proportional to class frequency [1, 2],
or the effective number [3].
- 𝑛𝑘: the number of samples in 𝑘-th class, 𝐶: the total number of classes.
- The generalized sampling probability for 𝑘-th class can be defined by
𝑞 𝑟, 𝑘 =
1/𝑛𝑘
𝑟
σ𝑘′=1
𝐶
1/𝑛𝑘′
𝑟
- As 𝑟 increases, weight of the minor class becomes increasingly larger than
that of the major class.
- Effective number[3] is defined as
𝐸 𝑘 =
1 − 𝛽𝑛𝑘
1 − 𝛽
8
The Majority Can Help the Minority: Context-rich Minority Oversampling for Long-tailed Classification
[1] Learning deep representation for imbalanced classification, CVPR, 2016.
[2] Exploring the limits of weakly supervised pretraining, ECCV, 2018.
[3] Class-balanced loss based on effective number of samples, CVPR, 2019.
𝑟 = 1
𝑟 = 2
Original
𝑟 = 0
Data distribution
Proposed Method: Algorithm
Introduction Proposed Method Experiment Conclusion
Algorithm
9
The Majority Can Help the Minority: Context-rich Minority Oversampling for Long-tailed Classification
Experimental Results
Introduction Proposed Method Experiment Conclusion
1. Datasets
◦Synthetic data:
▪CIFAR-100-LT (100 classes), ImageNet-LT (1,000 classes)
◦Real-world data:
▪iNaturalist 2018 (8,142 classes)
※ imbalance ratio: the ratio between the most frequent class and the least frequent class.
2. Evaluation metrics
◦Top-1 accuracy
◦Accuracy for disjoint sets (Many > 100, 20<=Med<=100, Few<20) [1]
10
The Majority Can Help the Minority: Context-rich Minority Oversampling for Long-tailed Classification
[1] Large-scale long-tailed recognition in an open world, CVPR, 2019.
Long-tailed classification benchmarks (ImageNet-LT)
Introduction Proposed Method Experiment Conclusion
1. Comparison with state-of-the-arts
11
The Majority Can Help the Minority: Context-rich Minority Oversampling for Long-tailed Classification
2. Comparison with oversampling methods
3. Results of longer training epochs
Analysis
Introduction Proposed Method Experiment Conclusion
1. Impact of different Q distributions
12
The Majority Can Help the Minority: Context-rich Minority Oversampling for Long-tailed Classification
2. Using different augmentation methods
3. Variants of CMO 4. Generated Images
Conclusion
Introduction Proposed Method Experiment Conclusion
✓We propose a novel context-rich minority oversampling that leverages the rich context of the majority
classes as background images.
✓It requires little additional computational cost and can be easily integrated into existing methods.
✓It is simple but effective that achieves the state-of-the-art performance.
✓We empirically prove the effectiveness of the proposed oversampling method through extensive
experiments and ablation studies.
13
The Majority Can Help the Minority: Context-rich Minority Oversampling for Long-tailed Classification
Conclusion
Introduction Proposed Method Experiment Conclusion
✓We propose a novel context-rich minority oversampling that leverages the rich context of the majority
classes as background images.
✓It requires little additional computational cost and can be easily integrated into existing methods.
✓It is simple but effective that achieves the state-of-the-art performance.
✓We empirically prove the effectiveness of the proposed oversampling method through extensive
experiments and ablation studies.
14
The Majority Can Help the Minority: Context-rich Minority Oversampling for Long-tailed Classification
Thank you!
Contact: seulki.park@snu.ac.kr
Code: https://github.com/naver-ai/cmo

More Related Content

What's hot

3a data link layer
3a data link layer 3a data link layer
3a data link layer
kavish dani
 
Clustering
ClusteringClustering
Over fitting underfitting
Over fitting underfittingOver fitting underfitting
Over fitting underfitting
SivapriyaS12
 
Decoding BCH-Code.pdf
Decoding BCH-Code.pdfDecoding BCH-Code.pdf
Decoding BCH-Code.pdf
KundanSasi
 
Convolutional Neural Networks : Popular Architectures
Convolutional Neural Networks : Popular ArchitecturesConvolutional Neural Networks : Popular Architectures
Convolutional Neural Networks : Popular Architectures
ananth
 
Transfer Learning: An overview
Transfer Learning: An overviewTransfer Learning: An overview
Transfer Learning: An overview
jins0618
 
Arithmetic coding
Arithmetic codingArithmetic coding
Arithmetic coding
Gidey Leul
 
Color image processing Presentation
Color image processing PresentationColor image processing Presentation
Color image processing Presentation
Revanth Chimmani
 
Color Image Processing
Color Image ProcessingColor Image Processing
Color Image Processing
kiruthiammu
 
Overview of Convolutional Neural Networks
Overview of Convolutional Neural NetworksOverview of Convolutional Neural Networks
Overview of Convolutional Neural Networks
ananth
 
Image Processing: Spatial filters
Image Processing: Spatial filtersImage Processing: Spatial filters
Image Processing: Spatial filters
A B Shinde
 
1.arithmetic & logical operations
1.arithmetic & logical operations1.arithmetic & logical operations
1.arithmetic & logical operations
mukesh bhardwaj
 
Fuzzy Logic Seminar with Implementation
Fuzzy Logic Seminar with ImplementationFuzzy Logic Seminar with Implementation
Fuzzy Logic Seminar with Implementation
Bhaumik Parmar
 
10 -- Overfitting and Underfitting.pptx
10 -- Overfitting and Underfitting.pptx10 -- Overfitting and Underfitting.pptx
10 -- Overfitting and Underfitting.pptx
kpcp
 
7. data link layer error detection and correction codes - parity and checksum
7. data link layer   error detection and correction codes - parity and checksum7. data link layer   error detection and correction codes - parity and checksum
7. data link layer error detection and correction codes - parity and checksum
JAIGANESH SEKAR
 
Data Link Layer| Error Detection
Data Link Layer| Error DetectionData Link Layer| Error Detection
Data Link Layer| Error Detection
Taimoor Muzaffar Gondal
 
空間濾波 Spatial Filtering
空間濾波 Spatial Filtering空間濾波 Spatial Filtering
空間濾波 Spatial Filtering
YKLee3434
 
Convolutional Neural Network (CNN)
Convolutional Neural Network (CNN)Convolutional Neural Network (CNN)
Convolutional Neural Network (CNN)
Abdulrazak Zakieh
 
Convolutional Neural Networks (CNN)
Convolutional Neural Networks (CNN)Convolutional Neural Networks (CNN)
Convolutional Neural Networks (CNN)
Gaurav Mittal
 
ICASSP 2018 Tutorial: Generative Adversarial Network and its Applications to ...
ICASSP 2018 Tutorial: Generative Adversarial Network and its Applications to ...ICASSP 2018 Tutorial: Generative Adversarial Network and its Applications to ...
ICASSP 2018 Tutorial: Generative Adversarial Network and its Applications to ...
宏毅 李
 

What's hot (20)

3a data link layer
3a data link layer 3a data link layer
3a data link layer
 
Clustering
ClusteringClustering
Clustering
 
Over fitting underfitting
Over fitting underfittingOver fitting underfitting
Over fitting underfitting
 
Decoding BCH-Code.pdf
Decoding BCH-Code.pdfDecoding BCH-Code.pdf
Decoding BCH-Code.pdf
 
Convolutional Neural Networks : Popular Architectures
Convolutional Neural Networks : Popular ArchitecturesConvolutional Neural Networks : Popular Architectures
Convolutional Neural Networks : Popular Architectures
 
Transfer Learning: An overview
Transfer Learning: An overviewTransfer Learning: An overview
Transfer Learning: An overview
 
Arithmetic coding
Arithmetic codingArithmetic coding
Arithmetic coding
 
Color image processing Presentation
Color image processing PresentationColor image processing Presentation
Color image processing Presentation
 
Color Image Processing
Color Image ProcessingColor Image Processing
Color Image Processing
 
Overview of Convolutional Neural Networks
Overview of Convolutional Neural NetworksOverview of Convolutional Neural Networks
Overview of Convolutional Neural Networks
 
Image Processing: Spatial filters
Image Processing: Spatial filtersImage Processing: Spatial filters
Image Processing: Spatial filters
 
1.arithmetic & logical operations
1.arithmetic & logical operations1.arithmetic & logical operations
1.arithmetic & logical operations
 
Fuzzy Logic Seminar with Implementation
Fuzzy Logic Seminar with ImplementationFuzzy Logic Seminar with Implementation
Fuzzy Logic Seminar with Implementation
 
10 -- Overfitting and Underfitting.pptx
10 -- Overfitting and Underfitting.pptx10 -- Overfitting and Underfitting.pptx
10 -- Overfitting and Underfitting.pptx
 
7. data link layer error detection and correction codes - parity and checksum
7. data link layer   error detection and correction codes - parity and checksum7. data link layer   error detection and correction codes - parity and checksum
7. data link layer error detection and correction codes - parity and checksum
 
Data Link Layer| Error Detection
Data Link Layer| Error DetectionData Link Layer| Error Detection
Data Link Layer| Error Detection
 
空間濾波 Spatial Filtering
空間濾波 Spatial Filtering空間濾波 Spatial Filtering
空間濾波 Spatial Filtering
 
Convolutional Neural Network (CNN)
Convolutional Neural Network (CNN)Convolutional Neural Network (CNN)
Convolutional Neural Network (CNN)
 
Convolutional Neural Networks (CNN)
Convolutional Neural Networks (CNN)Convolutional Neural Networks (CNN)
Convolutional Neural Networks (CNN)
 
ICASSP 2018 Tutorial: Generative Adversarial Network and its Applications to ...
ICASSP 2018 Tutorial: Generative Adversarial Network and its Applications to ...ICASSP 2018 Tutorial: Generative Adversarial Network and its Applications to ...
ICASSP 2018 Tutorial: Generative Adversarial Network and its Applications to ...
 

Similar to [CVPR 22] Context-rich Minority Oversampling for Long-tailed Classification

IMBALANCED DATA LEARNING APPROACHES REVIEW
IMBALANCED DATA LEARNING APPROACHES REVIEWIMBALANCED DATA LEARNING APPROACHES REVIEW
IMBALANCED DATA LEARNING APPROACHES REVIEW
IJDKP
 
Multi-Cluster Based Approach for skewed Data in Data Mining
Multi-Cluster Based Approach for skewed Data in Data MiningMulti-Cluster Based Approach for skewed Data in Data Mining
Multi-Cluster Based Approach for skewed Data in Data Mining
IOSR Journals
 
PROBABILITY BASED CLUSTER EXPANSION OVERSAMPLING TECHNIQUE FOR IMBALANCED DATA
PROBABILITY BASED CLUSTER EXPANSION OVERSAMPLING TECHNIQUE FOR IMBALANCED DATAPROBABILITY BASED CLUSTER EXPANSION OVERSAMPLING TECHNIQUE FOR IMBALANCED DATA
PROBABILITY BASED CLUSTER EXPANSION OVERSAMPLING TECHNIQUE FOR IMBALANCED DATA
cscpconf
 
Ijetr021251
Ijetr021251Ijetr021251
6145-Article Text-9370-1-10-20200513.pdf
6145-Article Text-9370-1-10-20200513.pdf6145-Article Text-9370-1-10-20200513.pdf
6145-Article Text-9370-1-10-20200513.pdf
chalachew5
 
When deep learners change their mind learning dynamics for active learning
When deep learners change their mind  learning dynamics for active learningWhen deep learners change their mind  learning dynamics for active learning
When deep learners change their mind learning dynamics for active learning
Devansh16
 
Analysis of Imbalanced Classification Algorithms A Perspective View
Analysis of Imbalanced Classification Algorithms A Perspective ViewAnalysis of Imbalanced Classification Algorithms A Perspective View
Analysis of Imbalanced Classification Algorithms A Perspective View
ijtsrd
 
Ijetr021252
Ijetr021252Ijetr021252
Lecture 6: Ensemble Methods
Lecture 6: Ensemble Methods Lecture 6: Ensemble Methods
Lecture 6: Ensemble Methods
Marina Santini
 
Learning On The Border:Active Learning in Imbalanced classification Data
Learning On The Border:Active Learning in Imbalanced classification DataLearning On The Border:Active Learning in Imbalanced classification Data
Learning On The Border:Active Learning in Imbalanced classification Data
萍華 楊
 
Ensemble learning
Ensemble learningEnsemble learning
Ensemble learning
Mustafa Sherazi
 
(Machine Learning) Ensemble learning
(Machine Learning) Ensemble learning (Machine Learning) Ensemble learning
(Machine Learning) Ensemble learning
Omkar Rane
 
An overview on data mining designed for imbalanced datasets
An overview on data mining designed for imbalanced datasetsAn overview on data mining designed for imbalanced datasets
An overview on data mining designed for imbalanced datasets
eSAT Publishing House
 
An overview on data mining designed for imbalanced datasets
An overview on data mining designed for imbalanced datasetsAn overview on data mining designed for imbalanced datasets
An overview on data mining designed for imbalanced datasets
eSAT Journals
 
Multilevel techniques for the clustering problem
Multilevel techniques for the clustering problemMultilevel techniques for the clustering problem
Multilevel techniques for the clustering problem
csandit
 
METHODS FOR INCREMENTAL LEARNING: A SURVEY
METHODS FOR INCREMENTAL LEARNING: A SURVEYMETHODS FOR INCREMENTAL LEARNING: A SURVEY
METHODS FOR INCREMENTAL LEARNING: A SURVEY
IJDKP
 
An approach for improved students’ performance prediction using homogeneous ...
An approach for improved students’ performance prediction  using homogeneous ...An approach for improved students’ performance prediction  using homogeneous ...
An approach for improved students’ performance prediction using homogeneous ...
IJECEIAES
 
Dnn guidelines
Dnn guidelinesDnn guidelines
Dnn guidelines
Naitik Shukla
 
COMBINED CLASSIFIERS FOR TIME SERIES SHAPELETS
COMBINED CLASSIFIERS FOR TIME SERIES SHAPELETSCOMBINED CLASSIFIERS FOR TIME SERIES SHAPELETS
COMBINED CLASSIFIERS FOR TIME SERIES SHAPELETS
cscpconf
 
COMBINED CLASSIFIERS FOR TIME SERIES SHAPELETS
COMBINED CLASSIFIERS FOR TIME SERIES SHAPELETSCOMBINED CLASSIFIERS FOR TIME SERIES SHAPELETS
COMBINED CLASSIFIERS FOR TIME SERIES SHAPELETS
csandit
 

Similar to [CVPR 22] Context-rich Minority Oversampling for Long-tailed Classification (20)

IMBALANCED DATA LEARNING APPROACHES REVIEW
IMBALANCED DATA LEARNING APPROACHES REVIEWIMBALANCED DATA LEARNING APPROACHES REVIEW
IMBALANCED DATA LEARNING APPROACHES REVIEW
 
Multi-Cluster Based Approach for skewed Data in Data Mining
Multi-Cluster Based Approach for skewed Data in Data MiningMulti-Cluster Based Approach for skewed Data in Data Mining
Multi-Cluster Based Approach for skewed Data in Data Mining
 
PROBABILITY BASED CLUSTER EXPANSION OVERSAMPLING TECHNIQUE FOR IMBALANCED DATA
PROBABILITY BASED CLUSTER EXPANSION OVERSAMPLING TECHNIQUE FOR IMBALANCED DATAPROBABILITY BASED CLUSTER EXPANSION OVERSAMPLING TECHNIQUE FOR IMBALANCED DATA
PROBABILITY BASED CLUSTER EXPANSION OVERSAMPLING TECHNIQUE FOR IMBALANCED DATA
 
Ijetr021251
Ijetr021251Ijetr021251
Ijetr021251
 
6145-Article Text-9370-1-10-20200513.pdf
6145-Article Text-9370-1-10-20200513.pdf6145-Article Text-9370-1-10-20200513.pdf
6145-Article Text-9370-1-10-20200513.pdf
 
When deep learners change their mind learning dynamics for active learning
When deep learners change their mind  learning dynamics for active learningWhen deep learners change their mind  learning dynamics for active learning
When deep learners change their mind learning dynamics for active learning
 
Analysis of Imbalanced Classification Algorithms A Perspective View
Analysis of Imbalanced Classification Algorithms A Perspective ViewAnalysis of Imbalanced Classification Algorithms A Perspective View
Analysis of Imbalanced Classification Algorithms A Perspective View
 
Ijetr021252
Ijetr021252Ijetr021252
Ijetr021252
 
Lecture 6: Ensemble Methods
Lecture 6: Ensemble Methods Lecture 6: Ensemble Methods
Lecture 6: Ensemble Methods
 
Learning On The Border:Active Learning in Imbalanced classification Data
Learning On The Border:Active Learning in Imbalanced classification DataLearning On The Border:Active Learning in Imbalanced classification Data
Learning On The Border:Active Learning in Imbalanced classification Data
 
Ensemble learning
Ensemble learningEnsemble learning
Ensemble learning
 
(Machine Learning) Ensemble learning
(Machine Learning) Ensemble learning (Machine Learning) Ensemble learning
(Machine Learning) Ensemble learning
 
An overview on data mining designed for imbalanced datasets
An overview on data mining designed for imbalanced datasetsAn overview on data mining designed for imbalanced datasets
An overview on data mining designed for imbalanced datasets
 
An overview on data mining designed for imbalanced datasets
An overview on data mining designed for imbalanced datasetsAn overview on data mining designed for imbalanced datasets
An overview on data mining designed for imbalanced datasets
 
Multilevel techniques for the clustering problem
Multilevel techniques for the clustering problemMultilevel techniques for the clustering problem
Multilevel techniques for the clustering problem
 
METHODS FOR INCREMENTAL LEARNING: A SURVEY
METHODS FOR INCREMENTAL LEARNING: A SURVEYMETHODS FOR INCREMENTAL LEARNING: A SURVEY
METHODS FOR INCREMENTAL LEARNING: A SURVEY
 
An approach for improved students’ performance prediction using homogeneous ...
An approach for improved students’ performance prediction  using homogeneous ...An approach for improved students’ performance prediction  using homogeneous ...
An approach for improved students’ performance prediction using homogeneous ...
 
Dnn guidelines
Dnn guidelinesDnn guidelines
Dnn guidelines
 
COMBINED CLASSIFIERS FOR TIME SERIES SHAPELETS
COMBINED CLASSIFIERS FOR TIME SERIES SHAPELETSCOMBINED CLASSIFIERS FOR TIME SERIES SHAPELETS
COMBINED CLASSIFIERS FOR TIME SERIES SHAPELETS
 
COMBINED CLASSIFIERS FOR TIME SERIES SHAPELETS
COMBINED CLASSIFIERS FOR TIME SERIES SHAPELETSCOMBINED CLASSIFIERS FOR TIME SERIES SHAPELETS
COMBINED CLASSIFIERS FOR TIME SERIES SHAPELETS
 

Recently uploaded

Finale of the Year: Apply for Next One!
Finale of the Year: Apply for Next One!Finale of the Year: Apply for Next One!
Finale of the Year: Apply for Next One!
GDSC PJATK
 
Main news related to the CCS TSI 2023 (2023/1695)
Main news related to the CCS TSI 2023 (2023/1695)Main news related to the CCS TSI 2023 (2023/1695)
Main news related to the CCS TSI 2023 (2023/1695)
Jakub Marek
 
Building Production Ready Search Pipelines with Spark and Milvus
Building Production Ready Search Pipelines with Spark and MilvusBuilding Production Ready Search Pipelines with Spark and Milvus
Building Production Ready Search Pipelines with Spark and Milvus
Zilliz
 
5th LF Energy Power Grid Model Meet-up Slides
5th LF Energy Power Grid Model Meet-up Slides5th LF Energy Power Grid Model Meet-up Slides
5th LF Energy Power Grid Model Meet-up Slides
DanBrown980551
 
HCL Notes und Domino Lizenzkostenreduzierung in der Welt von DLAU
HCL Notes und Domino Lizenzkostenreduzierung in der Welt von DLAUHCL Notes und Domino Lizenzkostenreduzierung in der Welt von DLAU
HCL Notes und Domino Lizenzkostenreduzierung in der Welt von DLAU
panagenda
 
Fueling AI with Great Data with Airbyte Webinar
Fueling AI with Great Data with Airbyte WebinarFueling AI with Great Data with Airbyte Webinar
Fueling AI with Great Data with Airbyte Webinar
Zilliz
 
Unlock the Future of Search with MongoDB Atlas_ Vector Search Unleashed.pdf
Unlock the Future of Search with MongoDB Atlas_ Vector Search Unleashed.pdfUnlock the Future of Search with MongoDB Atlas_ Vector Search Unleashed.pdf
Unlock the Future of Search with MongoDB Atlas_ Vector Search Unleashed.pdf
Malak Abu Hammad
 
WeTestAthens: Postman's AI & Automation Techniques
WeTestAthens: Postman's AI & Automation TechniquesWeTestAthens: Postman's AI & Automation Techniques
WeTestAthens: Postman's AI & Automation Techniques
Postman
 
Trusted Execution Environment for Decentralized Process Mining
Trusted Execution Environment for Decentralized Process MiningTrusted Execution Environment for Decentralized Process Mining
Trusted Execution Environment for Decentralized Process Mining
LucaBarbaro3
 
HCL Notes and Domino License Cost Reduction in the World of DLAU
HCL Notes and Domino License Cost Reduction in the World of DLAUHCL Notes and Domino License Cost Reduction in the World of DLAU
HCL Notes and Domino License Cost Reduction in the World of DLAU
panagenda
 
Generating privacy-protected synthetic data using Secludy and Milvus
Generating privacy-protected synthetic data using Secludy and MilvusGenerating privacy-protected synthetic data using Secludy and Milvus
Generating privacy-protected synthetic data using Secludy and Milvus
Zilliz
 
UI5 Controls simplified - UI5con2024 presentation
UI5 Controls simplified - UI5con2024 presentationUI5 Controls simplified - UI5con2024 presentation
UI5 Controls simplified - UI5con2024 presentation
Wouter Lemaire
 
Azure API Management to expose backend services securely
Azure API Management to expose backend services securelyAzure API Management to expose backend services securely
Azure API Management to expose backend services securely
Dinusha Kumarasiri
 
Letter and Document Automation for Bonterra Impact Management (fka Social Sol...
Letter and Document Automation for Bonterra Impact Management (fka Social Sol...Letter and Document Automation for Bonterra Impact Management (fka Social Sol...
Letter and Document Automation for Bonterra Impact Management (fka Social Sol...
Jeffrey Haguewood
 
leewayhertz.com-AI in predictive maintenance Use cases technologies benefits ...
leewayhertz.com-AI in predictive maintenance Use cases technologies benefits ...leewayhertz.com-AI in predictive maintenance Use cases technologies benefits ...
leewayhertz.com-AI in predictive maintenance Use cases technologies benefits ...
alexjohnson7307
 
A Comprehensive Guide to DeFi Development Services in 2024
A Comprehensive Guide to DeFi Development Services in 2024A Comprehensive Guide to DeFi Development Services in 2024
A Comprehensive Guide to DeFi Development Services in 2024
Intelisync
 
Taking AI to the Next Level in Manufacturing.pdf
Taking AI to the Next Level in Manufacturing.pdfTaking AI to the Next Level in Manufacturing.pdf
Taking AI to the Next Level in Manufacturing.pdf
ssuserfac0301
 
Let's Integrate MuleSoft RPA, COMPOSER, APM with AWS IDP along with Slack
Let's Integrate MuleSoft RPA, COMPOSER, APM with AWS IDP along with SlackLet's Integrate MuleSoft RPA, COMPOSER, APM with AWS IDP along with Slack
Let's Integrate MuleSoft RPA, COMPOSER, APM with AWS IDP along with Slack
shyamraj55
 
Energy Efficient Video Encoding for Cloud and Edge Computing Instances
Energy Efficient Video Encoding for Cloud and Edge Computing InstancesEnergy Efficient Video Encoding for Cloud and Edge Computing Instances
Energy Efficient Video Encoding for Cloud and Edge Computing Instances
Alpen-Adria-Universität
 
TrustArc Webinar - 2024 Global Privacy Survey
TrustArc Webinar - 2024 Global Privacy SurveyTrustArc Webinar - 2024 Global Privacy Survey
TrustArc Webinar - 2024 Global Privacy Survey
TrustArc
 

Recently uploaded (20)

Finale of the Year: Apply for Next One!
Finale of the Year: Apply for Next One!Finale of the Year: Apply for Next One!
Finale of the Year: Apply for Next One!
 
Main news related to the CCS TSI 2023 (2023/1695)
Main news related to the CCS TSI 2023 (2023/1695)Main news related to the CCS TSI 2023 (2023/1695)
Main news related to the CCS TSI 2023 (2023/1695)
 
Building Production Ready Search Pipelines with Spark and Milvus
Building Production Ready Search Pipelines with Spark and MilvusBuilding Production Ready Search Pipelines with Spark and Milvus
Building Production Ready Search Pipelines with Spark and Milvus
 
5th LF Energy Power Grid Model Meet-up Slides
5th LF Energy Power Grid Model Meet-up Slides5th LF Energy Power Grid Model Meet-up Slides
5th LF Energy Power Grid Model Meet-up Slides
 
HCL Notes und Domino Lizenzkostenreduzierung in der Welt von DLAU
HCL Notes und Domino Lizenzkostenreduzierung in der Welt von DLAUHCL Notes und Domino Lizenzkostenreduzierung in der Welt von DLAU
HCL Notes und Domino Lizenzkostenreduzierung in der Welt von DLAU
 
Fueling AI with Great Data with Airbyte Webinar
Fueling AI with Great Data with Airbyte WebinarFueling AI with Great Data with Airbyte Webinar
Fueling AI with Great Data with Airbyte Webinar
 
Unlock the Future of Search with MongoDB Atlas_ Vector Search Unleashed.pdf
Unlock the Future of Search with MongoDB Atlas_ Vector Search Unleashed.pdfUnlock the Future of Search with MongoDB Atlas_ Vector Search Unleashed.pdf
Unlock the Future of Search with MongoDB Atlas_ Vector Search Unleashed.pdf
 
WeTestAthens: Postman's AI & Automation Techniques
WeTestAthens: Postman's AI & Automation TechniquesWeTestAthens: Postman's AI & Automation Techniques
WeTestAthens: Postman's AI & Automation Techniques
 
Trusted Execution Environment for Decentralized Process Mining
Trusted Execution Environment for Decentralized Process MiningTrusted Execution Environment for Decentralized Process Mining
Trusted Execution Environment for Decentralized Process Mining
 
HCL Notes and Domino License Cost Reduction in the World of DLAU
HCL Notes and Domino License Cost Reduction in the World of DLAUHCL Notes and Domino License Cost Reduction in the World of DLAU
HCL Notes and Domino License Cost Reduction in the World of DLAU
 
Generating privacy-protected synthetic data using Secludy and Milvus
Generating privacy-protected synthetic data using Secludy and MilvusGenerating privacy-protected synthetic data using Secludy and Milvus
Generating privacy-protected synthetic data using Secludy and Milvus
 
UI5 Controls simplified - UI5con2024 presentation
UI5 Controls simplified - UI5con2024 presentationUI5 Controls simplified - UI5con2024 presentation
UI5 Controls simplified - UI5con2024 presentation
 
Azure API Management to expose backend services securely
Azure API Management to expose backend services securelyAzure API Management to expose backend services securely
Azure API Management to expose backend services securely
 
Letter and Document Automation for Bonterra Impact Management (fka Social Sol...
Letter and Document Automation for Bonterra Impact Management (fka Social Sol...Letter and Document Automation for Bonterra Impact Management (fka Social Sol...
Letter and Document Automation for Bonterra Impact Management (fka Social Sol...
 
leewayhertz.com-AI in predictive maintenance Use cases technologies benefits ...
leewayhertz.com-AI in predictive maintenance Use cases technologies benefits ...leewayhertz.com-AI in predictive maintenance Use cases technologies benefits ...
leewayhertz.com-AI in predictive maintenance Use cases technologies benefits ...
 
A Comprehensive Guide to DeFi Development Services in 2024
A Comprehensive Guide to DeFi Development Services in 2024A Comprehensive Guide to DeFi Development Services in 2024
A Comprehensive Guide to DeFi Development Services in 2024
 
Taking AI to the Next Level in Manufacturing.pdf
Taking AI to the Next Level in Manufacturing.pdfTaking AI to the Next Level in Manufacturing.pdf
Taking AI to the Next Level in Manufacturing.pdf
 
Let's Integrate MuleSoft RPA, COMPOSER, APM with AWS IDP along with Slack
Let's Integrate MuleSoft RPA, COMPOSER, APM with AWS IDP along with SlackLet's Integrate MuleSoft RPA, COMPOSER, APM with AWS IDP along with Slack
Let's Integrate MuleSoft RPA, COMPOSER, APM with AWS IDP along with Slack
 
Energy Efficient Video Encoding for Cloud and Edge Computing Instances
Energy Efficient Video Encoding for Cloud and Edge Computing InstancesEnergy Efficient Video Encoding for Cloud and Edge Computing Instances
Energy Efficient Video Encoding for Cloud and Edge Computing Instances
 
TrustArc Webinar - 2024 Global Privacy Survey
TrustArc Webinar - 2024 Global Privacy SurveyTrustArc Webinar - 2024 Global Privacy Survey
TrustArc Webinar - 2024 Global Privacy Survey
 

[CVPR 22] Context-rich Minority Oversampling for Long-tailed Classification

  • 1. The Majority Can Help the Minority: Context-rich Minority Oversampling for Long-tailed Classification 1Seoul National University, 2NAVER AI Lab Poster ID: 159a & Poster Time: 22, Jun. 10:00-12:30 Seulki Park1 Byeongho Heo2 Sangdoo Yun2 Jin Young Choi1 Youngkyu Hong2
  • 2. Long-tailed Classification 2 Introduction Proposed Method Experiment Conclusion Many real-world data often exhibit long-tailed distribution. ✓ The model trained on such imbalanced data tends to overfit the majority classes. ✓ That is, the model performs poorly on minority classes. Problem Definition: ● Input: Long-tailed (imbalanced) training data & uniform-distributed (balanced) test data. ● Goal: To make a robust model that can generalize well on balanced test data. The Majority Can Help the Minority: Context-rich Minority Oversampling for Long-tailed Classification Faces (Zhang et al., 2017) Places (Wang et al., 2017) Species (Van Horn et al., 2018) Actions (Zhang et al., 2019) * Images by authors.
  • 3. Previous Oversampling Methods Introduction Proposed Method Experiment Conclusion 1. Random Oversampling (ROS) ◦ A simple and straightforward method which repeatedly oversample minor classes. ◦ However, this may intensify overfitting problem [1]. 2. Synthetic Minority Over-sampling Technique (SMOTE), 2002 ◦ Oversamples minority samples by interpolating between existing minority samples and their nearest minority neighbors. ◦ However, difficulties for end-to-end algorithm and largescale image datasets due to the high computational complexity of calculating K-Nearest Neighbor for every sample. 3 The Majority Can Help the Minority: Context-rich Minority Oversampling for Long-tailed Classification [1] Deep imbalanced attribute classification using visual attention aggregation, ECCV, 2018. SMOTE Figure from: http://www.incodom.kr/SMOTE
  • 4. Previous Oversampling Methods Introduction Proposed Method Experiment Conclusion 3. Generative Adversarial Minority Oversampling (GAMO), 2019 ◦ Produces new minority samples by training a convex generator, inspired by the success of generative adversarial networks (GANs). ◦ However, difficult to train a generator (mode collapse) & additional training cost. 4. MetaSAug: Meta Semantic Augmentation for Long-Tailed Visual Recognition, 2021 ◦ Uses implicit semantic data augmentation (ISDA) algorithm [1]. ◦ However, this meta-learning-based method requires additional balanced validation, and hundreds and thousands of iterations for training (high training cost). 4 The Majority Can Help the Minority: Context-rich Minority Oversampling for Long-tailed Classification [1] Implicit semantic data augmentation for deep networks, NeurIPS, 2019.
  • 5. Previous Oversampling Methods Introduction Proposed Method Experiment Conclusion Random Oversampling ◦ A simple and straightforward method which repeatedly oversample minor classes. ◦ However, this may intensify overfitting problem [1]. Synthetic Minority Over-sampling Technique (SMOTE), 2002 ◦ Oversamples minority samples by interpolating between existing minority samples and their nearest minority neighbors. ◦ However, difficulties for end-to-end algorithm and largescale image datasets due to the high computational complexity of calculating K-Nearest Neighbor for every sample. Generative Adversarial Minority Oversampling (GAMO), 2019 ◦ Produces new minority samples by training a convex generator, inspired by the success of generative adversarial networks (GANs). ◦ However, difficult to train a generator (mode collapse) & additional training cost. MetaSAug: Meta Semantic Augmentation for Long-Tailed Visual Recognition, 2021 ◦ Uses implicit semantic data augmentation (ISDA) algorithm [1]. ◦ However, this meta-learning-based method requires additional balanced validation, and hundreds and thousands of iterations for training (high training cost). 5 The Majority Can Help the Minority: Context-rich Minority Oversampling for Long-tailed Classification Limitation of previous methods: 1) Simple methods generate only context-limited images. (e.g, ROS, SMOTE) - Limited improvement especially when imbalance is severe. 2) Recent methods require additional expensive training cost (e.g, GAMO, MetaSAug) - E.g., training generator, additional balanced validation set, longer training epochs. → We need ‘Simple & Context-rich’ oversampling method!
  • 6. Motivation Introduction Proposed Method Experiment Conclusion Q. How can we generate diverse ‘context-rich minority samples’ from long-tailed distribution? A. Let’s pay attention to the characteristics of long-tailed distributions. Key Observations: ✓ Majority class samples are data-rich and information-rich! → Let’s use the affluent information of the majority samples to generate new minority samples. Key Idea - We can use the rich major-class images as the background for the newly created minor-class images. 6 The Majority Can Help the Minority: Context-rich Minority Oversampling for Long-tailed Classification
  • 7. Proposed Method: Context-rich Minority Oversampling (CMO) Introduction Proposed Method Experiment Conclusion Recap: CutMix (Yun et al., 2019) - A simple but effective data augmentation method used in many visual tasks. ෤ 𝑥 = 𝑴⨀𝑥𝑏 + 𝟏 − 𝑴 ⨀𝑥𝑓 , ෤ 𝑦 = 𝜆𝑦𝑏 + 1 − 𝜆 𝑦𝑓 𝑥𝑏 , 𝑦𝑏 , 𝑥𝑓 , 𝑦𝑓 ~ 𝑃 𝑴 ∈ 0, 1 𝑊×𝐻 : a binary mask → designed for a class balanced dataset. Naively using CutMix generates more samples of the majority classes. Context-rich Minority Oversampling (CMO) - For an imbalanced dataset, we use different distributions for background and foreground images. 𝑥𝑏 , 𝑦𝑏 ~ 𝑃, 𝑥𝑓 , 𝑦𝑓 ~ 𝑄 𝑄 : minor-class-weighted distribution. 7 The Majority Can Help the Minority: Context-rich Minority Oversampling for Long-tailed Classification CutMix Comparison with CutMix [3]
  • 8. Proposed Method: Minor-class-weighted distribution Introduction Proposed Method Experiment Conclusion How to design minor-class-weighted sampling strategies? - Re-weighting methods have provided a way how to assign appropriate weights to samples. - Commonly used sampling strategies give a weight inversely proportional to class frequency [1, 2], or the effective number [3]. - 𝑛𝑘: the number of samples in 𝑘-th class, 𝐶: the total number of classes. - The generalized sampling probability for 𝑘-th class can be defined by 𝑞 𝑟, 𝑘 = 1/𝑛𝑘 𝑟 σ𝑘′=1 𝐶 1/𝑛𝑘′ 𝑟 - As 𝑟 increases, weight of the minor class becomes increasingly larger than that of the major class. - Effective number[3] is defined as 𝐸 𝑘 = 1 − 𝛽𝑛𝑘 1 − 𝛽 8 The Majority Can Help the Minority: Context-rich Minority Oversampling for Long-tailed Classification [1] Learning deep representation for imbalanced classification, CVPR, 2016. [2] Exploring the limits of weakly supervised pretraining, ECCV, 2018. [3] Class-balanced loss based on effective number of samples, CVPR, 2019. 𝑟 = 1 𝑟 = 2 Original 𝑟 = 0 Data distribution
  • 9. Proposed Method: Algorithm Introduction Proposed Method Experiment Conclusion Algorithm 9 The Majority Can Help the Minority: Context-rich Minority Oversampling for Long-tailed Classification
  • 10. Experimental Results Introduction Proposed Method Experiment Conclusion 1. Datasets ◦Synthetic data: ▪CIFAR-100-LT (100 classes), ImageNet-LT (1,000 classes) ◦Real-world data: ▪iNaturalist 2018 (8,142 classes) ※ imbalance ratio: the ratio between the most frequent class and the least frequent class. 2. Evaluation metrics ◦Top-1 accuracy ◦Accuracy for disjoint sets (Many > 100, 20<=Med<=100, Few<20) [1] 10 The Majority Can Help the Minority: Context-rich Minority Oversampling for Long-tailed Classification [1] Large-scale long-tailed recognition in an open world, CVPR, 2019.
  • 11. Long-tailed classification benchmarks (ImageNet-LT) Introduction Proposed Method Experiment Conclusion 1. Comparison with state-of-the-arts 11 The Majority Can Help the Minority: Context-rich Minority Oversampling for Long-tailed Classification 2. Comparison with oversampling methods 3. Results of longer training epochs
  • 12. Analysis Introduction Proposed Method Experiment Conclusion 1. Impact of different Q distributions 12 The Majority Can Help the Minority: Context-rich Minority Oversampling for Long-tailed Classification 2. Using different augmentation methods 3. Variants of CMO 4. Generated Images
  • 13. Conclusion Introduction Proposed Method Experiment Conclusion ✓We propose a novel context-rich minority oversampling that leverages the rich context of the majority classes as background images. ✓It requires little additional computational cost and can be easily integrated into existing methods. ✓It is simple but effective that achieves the state-of-the-art performance. ✓We empirically prove the effectiveness of the proposed oversampling method through extensive experiments and ablation studies. 13 The Majority Can Help the Minority: Context-rich Minority Oversampling for Long-tailed Classification
  • 14. Conclusion Introduction Proposed Method Experiment Conclusion ✓We propose a novel context-rich minority oversampling that leverages the rich context of the majority classes as background images. ✓It requires little additional computational cost and can be easily integrated into existing methods. ✓It is simple but effective that achieves the state-of-the-art performance. ✓We empirically prove the effectiveness of the proposed oversampling method through extensive experiments and ablation studies. 14 The Majority Can Help the Minority: Context-rich Minority Oversampling for Long-tailed Classification Thank you! Contact: seulki.park@snu.ac.kr Code: https://github.com/naver-ai/cmo