Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

HiGDA: Hierarchical Graph of Nodes to Learn Local-to-Global Topology for Semi-Supervised Domain Adaptation

Ba Hung Ngo, Doanh C. Bui, Nhat-Tuong Do-Tran, Tae Jong Choi

TL;DR

HiGDA introduces a hierarchical graph of nodes to explicitly model local-to-global representations for semi-supervised domain adaptation. It combines a Local Graph (LoG) that reasoned over image patches with a Global Graph (GoG) that aggregates samples by category, augmented by Graph Active Learning (GAL) to utilize unlabeled target data. The approach achieves state-of-the-art results on Office-Home, DomainNet, and VisDA2017, and shows strong compatibility with existing SSDA methods such as MME and AAC. By explicitly modeling both patch-level structure and cross-sample category relations, HiGDA delivers compact, robust domain-aligned representations with interpretable graph-based reasoning.

Abstract

The enhanced representational power and broad applicability of deep learning models have attracted significant interest from the research community in recent years. However, these models often struggle to perform effectively under domain shift conditions, where the training data (the source domain) is related to but exhibits different distributions from the testing data (the target domain). To address this challenge, previous studies have attempted to reduce the domain gap between source and target data by incorporating a few labeled target samples during training - a technique known as semi-supervised domain adaptation (SSDA). While this strategy has demonstrated notable improvements in classification performance, the network architectures used in these approaches primarily focus on exploiting the features of individual images, leaving room for improvement in capturing rich representations. In this study, we introduce a Hierarchical Graph of Nodes designed to simultaneously present representations at both feature and category levels. At the feature level, we introduce a local graph to identify the most relevant patches within an image, facilitating adaptability to defined main object representations. At the category level, we employ a global graph to aggregate the features from samples within the same category, thereby enriching overall representations. Extensive experiments on widely used SSDA benchmark datasets, including Office-Home, DomainNet, and VisDA2017, demonstrate that both quantitative and qualitative results substantiate the effectiveness of HiGDA, establishing it as a new state-of-the-art method.

HiGDA: Hierarchical Graph of Nodes to Learn Local-to-Global Topology for Semi-Supervised Domain Adaptation

TL;DR

HiGDA introduces a hierarchical graph of nodes to explicitly model local-to-global representations for semi-supervised domain adaptation. It combines a Local Graph (LoG) that reasoned over image patches with a Global Graph (GoG) that aggregates samples by category, augmented by Graph Active Learning (GAL) to utilize unlabeled target data. The approach achieves state-of-the-art results on Office-Home, DomainNet, and VisDA2017, and shows strong compatibility with existing SSDA methods such as MME and AAC. By explicitly modeling both patch-level structure and cross-sample category relations, HiGDA delivers compact, robust domain-aligned representations with interpretable graph-based reasoning.

Abstract

The enhanced representational power and broad applicability of deep learning models have attracted significant interest from the research community in recent years. However, these models often struggle to perform effectively under domain shift conditions, where the training data (the source domain) is related to but exhibits different distributions from the testing data (the target domain). To address this challenge, previous studies have attempted to reduce the domain gap between source and target data by incorporating a few labeled target samples during training - a technique known as semi-supervised domain adaptation (SSDA). While this strategy has demonstrated notable improvements in classification performance, the network architectures used in these approaches primarily focus on exploiting the features of individual images, leaving room for improvement in capturing rich representations. In this study, we introduce a Hierarchical Graph of Nodes designed to simultaneously present representations at both feature and category levels. At the feature level, we introduce a local graph to identify the most relevant patches within an image, facilitating adaptability to defined main object representations. At the category level, we employ a global graph to aggregate the features from samples within the same category, thereby enriching overall representations. Extensive experiments on widely used SSDA benchmark datasets, including Office-Home, DomainNet, and VisDA2017, demonstrate that both quantitative and qualitative results substantiate the effectiveness of HiGDA, establishing it as a new state-of-the-art method.

Paper Structure

This paper contains 29 sections, 13 equations, 4 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Methodology
  4. Problem definition
  5. Overview of hierarchical graph design
  6. Local graph.
  7. Global graph.
  8. HiGDA learning process
  9. Local-graph network.
  10. Global-graph network.
  11. Training scheme
  12. Supervised learning.
  13. Global edge supervision.
  14. Final objective function.
  15. Graph Active Learning Strategy
  16. ...and 14 more sections

Figures (4)

  • Figure 1: Overview of the hierarchical graph of nodes. Input images are considered global nodes, each comprising multiple sub-patches that are considered local nodes.
  • Figure 2: Overview of HiGDA. Each image forms a local graph $\mathcal{G}^{(L)}$, and a mini-batch forms a global graph $\mathcal{G}^{(G)}$ to explore feature to category level representations. Then, GAL minimizes bias from the source dataset.
  • Figure 3: GradCAM results extracted by the local graph. Please zoom in for viewing ease.
  • Figure 4: t-SNE tsne visualization on DomainNet of 10 classes in the real to sketch (rel$\rightarrow$skt) task. Please zoom in for viewing ease.