Enhancing Fine-grained Image Classification through Attentive Batch Training

2024-12-27

AI Summary

Fine-grained image classification suffers from difficulty in learning discriminative features due to high inter-class visual similarity. To address this, we propose the Relation-based Batch Integration (RBI) framework, introducing two novel components: Residual Relation Attention (RRA) and Relation Position Encoding (RPE). RBI is the first method to explicitly model structured intra-batch image relationships to guide feature enhancement—leveraging visual correlations among samples within the same mini-batch for discriminative representation learning. It enables a plug-and-play attention training paradigm without architectural modification. Extensive experiments demonstrate state-of-the-art performance: +2.78% accuracy on CUB200-2011, +3.83% on Stanford Dogs (achieving 95.79%), and 93.71% on Tiny-ImageNet—surpassing all prior methods on these benchmarks.

Technology Category

• Computer Vision: Representation Learning for Vision
• Reasoning under Uncertainty: Relational Probabilistic Models
• Machine Learning: Representation Learning

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Abstract

Fine-grained image classification, which is a challenging task in computer vision, requires precise differentiation among visually similar object categories. In this paper, we propose 1) a novel module called Residual Relationship Attention (RRA) that leverages the relationships between images within each training batch to effectively integrate visual feature vectors of batch images and 2) a novel technique called Relationship Position Encoding (RPE), which encodes the positions of relationships between original images in a batch and effectively preserves the relationship information between images within the batch. Additionally, we design a novel framework, namely Relationship Batch Integration (RBI), which utilizes RRA in conjunction with RPE, allowing the discernment of vital visual features that may remain elusive when examining a singular image representative of a particular class. Through extensive experiments, our proposed method demonstrates significant improvements in the accuracy of different fine-grained classifiers, with an average increase of $(+2.78%)$ and $(+3.83%)$ on the CUB200-2011 and Stanford Dog datasets, respectively, while achieving a state-of-the-art results $(95.79%)$ on the Stanford Dog dataset. Despite not achieving the same level of improvement as in fine-grained image classification, our method still demonstrates its prowess in leveraging general image classification by attaining a state-of-the-art result of $(93.71%)$ on the Tiny-Imagenet dataset. Furthermore, our method serves as a plug-in refinement module and can be easily integrated into different networks.

Problem

• Fine-grained Image Classification
• Computer Vision
• Accuracy Improvement

Innovation

• Residual Relational Attention (RRA)
• Relational Positional Encoding (RPE)
• Relational Batch Integration (RBI)