TasksSotADatasetsPapersMethodsSubmitAbout
Papers With Code 2

A community resource for machine learning research: papers, code, benchmarks, and state-of-the-art results.

Explore

Notable BenchmarksAll SotADatasetsPapersMethods

Community

Submit ResultsAbout

Data sourced from the PWC Archive (CC-BY-SA 4.0). Built by the community, for the community.

Papers/MetaSAug: Meta Semantic Augmentation for Long-Tailed Visua...

MetaSAug: Meta Semantic Augmentation for Long-Tailed Visual Recognition

Shuang Li, Kaixiong Gong, Chi Harold Liu, Yulin Wang, Feng Qiao, Xinjing Cheng

2021-03-23CVPR 2021 1Meta-LearningImage ClassificationLong-tail LearningData Augmentation
PaperPDFCode(official)

Abstract

Real-world training data usually exhibits long-tailed distribution, where several majority classes have a significantly larger number of samples than the remaining minority classes. This imbalance degrades the performance of typical supervised learning algorithms designed for balanced training sets. In this paper, we address this issue by augmenting minority classes with a recently proposed implicit semantic data augmentation (ISDA) algorithm, which produces diversified augmented samples by translating deep features along many semantically meaningful directions. Importantly, given that ISDA estimates the class-conditional statistics to obtain semantic directions, we find it ineffective to do this on minority classes due to the insufficient training data. To this end, we propose a novel approach to learn transformed semantic directions with meta-learning automatically. In specific, the augmentation strategy during training is dynamically optimized, aiming to minimize the loss on a small balanced validation set, which is approximated via a meta update step. Extensive empirical results on CIFAR-LT-10/100, ImageNet-LT, and iNaturalist 2017/2018 validate the effectiveness of our method.

Results

TaskDatasetMetricValueModel
Image ClassificationCIFAR-100-LT (ρ=200)Error Rate56.91MetaSAug-LDAM
Image ClassificationCIFAR-10-LT (ρ=10)Error Rate10.32MetaSAug-LDAM
Image ClassificationCIFAR-100-LT (ρ=50)Error Rate47.73MetaSAug-LDAM
Image ClassificationCIFAR-100-LT (ρ=10)Error Rate38.72MetaSAug-LDAM
Image ClassificationImageNet-LTTop-1 Accuracy50.03MetaSAug (ResNet-152)
Image ClassificationImageNet-LTTop-1 Accuracy47.39MetaSAug with CE loss
Image ClassificationCIFAR-10-LT (ρ=50)Error Rate15.66MetaSAug-LDAM
Image ClassificationCIFAR-100-LT (ρ=100)Error Rate51.99MetaSAug-LDAM
Image ClassificationCIFAR-10-LT (ρ=100)Error Rate19.34MetaSAug-LDAM
Image ClassificationCIFAR-10-LT (ρ=200)Error Rate22.65MetaSAug-LDAM
Few-Shot Image ClassificationCIFAR-100-LT (ρ=200)Error Rate56.91MetaSAug-LDAM
Few-Shot Image ClassificationCIFAR-10-LT (ρ=10)Error Rate10.32MetaSAug-LDAM
Few-Shot Image ClassificationCIFAR-100-LT (ρ=50)Error Rate47.73MetaSAug-LDAM
Few-Shot Image ClassificationCIFAR-100-LT (ρ=10)Error Rate38.72MetaSAug-LDAM
Few-Shot Image ClassificationImageNet-LTTop-1 Accuracy50.03MetaSAug (ResNet-152)
Few-Shot Image ClassificationImageNet-LTTop-1 Accuracy47.39MetaSAug with CE loss
Few-Shot Image ClassificationCIFAR-10-LT (ρ=50)Error Rate15.66MetaSAug-LDAM
Few-Shot Image ClassificationCIFAR-100-LT (ρ=100)Error Rate51.99MetaSAug-LDAM
Few-Shot Image ClassificationCIFAR-10-LT (ρ=100)Error Rate19.34MetaSAug-LDAM
Few-Shot Image ClassificationCIFAR-10-LT (ρ=200)Error Rate22.65MetaSAug-LDAM
Generalized Few-Shot ClassificationCIFAR-100-LT (ρ=200)Error Rate56.91MetaSAug-LDAM
Generalized Few-Shot ClassificationCIFAR-10-LT (ρ=10)Error Rate10.32MetaSAug-LDAM
Generalized Few-Shot ClassificationCIFAR-100-LT (ρ=50)Error Rate47.73MetaSAug-LDAM
Generalized Few-Shot ClassificationCIFAR-100-LT (ρ=10)Error Rate38.72MetaSAug-LDAM
Generalized Few-Shot ClassificationImageNet-LTTop-1 Accuracy50.03MetaSAug (ResNet-152)
Generalized Few-Shot ClassificationImageNet-LTTop-1 Accuracy47.39MetaSAug with CE loss
Generalized Few-Shot ClassificationCIFAR-10-LT (ρ=50)Error Rate15.66MetaSAug-LDAM
Generalized Few-Shot ClassificationCIFAR-100-LT (ρ=100)Error Rate51.99MetaSAug-LDAM
Generalized Few-Shot ClassificationCIFAR-10-LT (ρ=100)Error Rate19.34MetaSAug-LDAM
Generalized Few-Shot ClassificationCIFAR-10-LT (ρ=200)Error Rate22.65MetaSAug-LDAM
Long-tail LearningCIFAR-100-LT (ρ=200)Error Rate56.91MetaSAug-LDAM
Long-tail LearningCIFAR-10-LT (ρ=10)Error Rate10.32MetaSAug-LDAM
Long-tail LearningCIFAR-100-LT (ρ=50)Error Rate47.73MetaSAug-LDAM
Long-tail LearningCIFAR-100-LT (ρ=10)Error Rate38.72MetaSAug-LDAM
Long-tail LearningImageNet-LTTop-1 Accuracy50.03MetaSAug (ResNet-152)
Long-tail LearningImageNet-LTTop-1 Accuracy47.39MetaSAug with CE loss
Long-tail LearningCIFAR-10-LT (ρ=50)Error Rate15.66MetaSAug-LDAM
Long-tail LearningCIFAR-100-LT (ρ=100)Error Rate51.99MetaSAug-LDAM
Long-tail LearningCIFAR-10-LT (ρ=100)Error Rate19.34MetaSAug-LDAM
Long-tail LearningCIFAR-10-LT (ρ=200)Error Rate22.65MetaSAug-LDAM
Generalized Few-Shot LearningCIFAR-100-LT (ρ=200)Error Rate56.91MetaSAug-LDAM
Generalized Few-Shot LearningCIFAR-10-LT (ρ=10)Error Rate10.32MetaSAug-LDAM
Generalized Few-Shot LearningCIFAR-100-LT (ρ=50)Error Rate47.73MetaSAug-LDAM
Generalized Few-Shot LearningCIFAR-100-LT (ρ=10)Error Rate38.72MetaSAug-LDAM
Generalized Few-Shot LearningImageNet-LTTop-1 Accuracy50.03MetaSAug (ResNet-152)
Generalized Few-Shot LearningImageNet-LTTop-1 Accuracy47.39MetaSAug with CE loss
Generalized Few-Shot LearningCIFAR-10-LT (ρ=50)Error Rate15.66MetaSAug-LDAM
Generalized Few-Shot LearningCIFAR-100-LT (ρ=100)Error Rate51.99MetaSAug-LDAM
Generalized Few-Shot LearningCIFAR-10-LT (ρ=100)Error Rate19.34MetaSAug-LDAM
Generalized Few-Shot LearningCIFAR-10-LT (ρ=200)Error Rate22.65MetaSAug-LDAM

Related Papers

Automatic Classification and Segmentation of Tunnel Cracks Based on Deep Learning and Visual Explanations2025-07-18Adversarial attacks to image classification systems using evolutionary algorithms2025-07-17Efficient Adaptation of Pre-trained Vision Transformer underpinned by Approximately Orthogonal Fine-Tuning Strategy2025-07-17Federated Learning for Commercial Image Sources2025-07-17MUPAX: Multidimensional Problem Agnostic eXplainable AI2025-07-17Overview of the TalentCLEF 2025: Skill and Job Title Intelligence for Human Capital Management2025-07-17Pixel Perfect MegaMed: A Megapixel-Scale Vision-Language Foundation Model for Generating High Resolution Medical Images2025-07-17Are encoders able to learn landmarkers for warm-starting of Hyperparameter Optimization?2025-07-16