Leveraging Graph Neural Networks to Boost Fine-Grained Image Classification

Bao Quoc Bui, Duy Minh Le, Cong Tran, Anh Tuan Tran, Cuong Pham

Primary Area: representation learning for computer vision, audio, language, and other modalities

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Keywords: Fine-grained classification, Graph Neural Networks, post-hoc

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.

TL;DR: we introduce a novel approach that utilizes Graph Neural Network (GNN) blocks as a plug-in refinement module to enhance the performance of fine-grained classification

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 introduce a novel approach that utilizes Graph Neural Network (GNN) blocks to enhance the clustering capability of feature vectors extracted from images within a deep neural network (DNN) framework. These GNN blocks capture intricate dependencies between feature vectors by modeling them as nodes within a graph. This graph-based approach enables our model to learn contextual information and relationships that are essential for fine-grained categorization. In practice, 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 result $(95.79\%)$ on the Stanford Dog dataset. Furthermore, our method serves as a plug-in refinement module and can be easily integrated into different architectures.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 4939