Abstract:Conventional detectors suffer from performance degradation when dealing with long-tailed data due to a classification bias towards the majority head categories. In this paper, we contend that the learning bias originates from two factors: 1) the unequal competition arising from the imbalanced distribution of foreground categories, and 2) the lack of sample diversity in tail categories. To tackle these issues, we introduce a unified framework called BAlanced CLassification (BACL), which enables adaptive rectification of inequalities caused by disparities in category distribution and dynamic intensification of sample diversities in a synchronized manner. Specifically, a novel foreground classification balance loss (FCBL) is developed to ameliorate the domination of head categories and shift attention to difficult-to-differentiate categories by introducing pairwise class-aware margins and auto-adjusted weight terms, respectively. This loss prevents the over-suppression of tail categories in the context of unequal competition. Moreover, we propose a dynamic feature hallucination module (FHM), which enhances the representation of tail categories in the feature space by synthesizing hallucinated samples to introduce additional data variances. In this divide-and-conquer approach, BACL sets a new state-of-the-art on the challenging LVIS benchmark with a decoupled training pipeline, surpassing vanilla Faster R-CNN with ResNet-50-FPN by 5.8% AP and 16.1% AP for overall and tail categories. Extensive experiments demonstrate that BACL consistently achieves performance improvements across various datasets with different backbones and architectures. Code and models are available at <a class="link-external link-https" href="https://github.com/Tianhao-Qi/BACL" rel="external noopener nofollow">this https URL</a>.

Class Activation Maps-based Feature Augmentation for long-tailed classification

Feature Space Augmentation for Long-Tailed Data

Calibrating Class Activation Maps for Long-Tailed Visual Recognition

ECS-SC: Long-tailed classification via data augmentation based on easily confused sample selection and combination

SAFA: Sample-Adaptive Feature Augmentation for Long-Tailed Image Classification

Attentive Feature Augmentation for Long-Tailed Visual Recognition

Label-Specific Feature Augmentation for Long-Tailed Multi-Label Text Classification

Class Activation Map Generation by Representative Class Selection and Multi-Layer Feature Fusion

Instance-Specific Semantic Augmentation for Long-Tailed Image Classification

LayerCAM: Exploring Hierarchical Class Activation Maps for Localization

ProAug: Prototype-Based Augmentation for Long-Tailed Image Classification.

Exploring the auxiliary learning for long-tailed visual recognition

Extracting Class Activation Maps from Non-Discriminative Features as well

Text-guided Fourier Augmentation for long-tailed recognition

Cumulative Dual-Branch Network Framework for Long-Tailed Multi-Class Classification

Data-Centric Long-Tailed Image Recognition

Multi-view Feature Augmentation with Adaptive Class Activation Mapping

CAM-loss: Towards Learning Spatially Discriminative Feature Representations

Improving Long-Tailed Classification from Instance Level

Balanced Classification: A Unified Framework for Long-Tailed Object Detection

CUDA: Curriculum of Data Augmentation for Long-Tailed Recognition