Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A Source-Free Approach for Domain Adaptation via Multiview Image Transformation and Latent Space Consistency

Debopom Sutradhar, Md. Abdur Rahman, Mohaimenul Azam Khan Raiaan, Reem E. Mohamed, Sami Azam

TL;DR

The paper tackles domain adaptation without source data by introducing a source-free framework that learns domain-invariant features from the target domain through multiview augmentation and latent space consistency. A ConvNeXt-based encoder maps inputs to a latent space, where two augmented views of each target sample are enforced to have similar latent representations, via a combined loss $\mathcal{L} = \mathcal{L}_{class} + \lambda \mathcal{L}_{cons}$. Key contributions include the latent space consistency mechanism, a concrete two-view augmentation strategy, and empirical validation on Office-31, Office-Home, and Office-Caltech showing strong performance and robustness without adversarial training or pseudo-labeling. The approach offers a practical, efficient alternative for SFDA with potential impact on real-world deployment where access to source data is restricted. Overall, the method demonstrates competitive accuracy gains and consistent domain-invariant feature learning directly from the target domain.

Abstract

Domain adaptation (DA) addresses the challenge of transferring knowledge from a source domain to a target domain where image data distributions may differ. Existing DA methods often require access to source domain data, adversarial training, or complex pseudo-labeling techniques, which are computationally expensive. To address these challenges, this paper introduces a novel source-free domain adaptation method. It is the first approach to use multiview augmentation and latent space consistency techniques to learn domain-invariant features directly from the target domain. Our method eliminates the need for source-target alignment or pseudo-label refinement by learning transferable representations solely from the target domain by enforcing consistency between multiple augmented views in the latent space. Additionally, the method ensures consistency in the learned features by generating multiple augmented views of target domain data and minimizing the distance between their feature representations in the latent space. We also introduce a ConvNeXt-based encoder and design a loss function that combines classification and consistency objectives to drive effective adaptation directly from the target domain. The proposed model achieves an average classification accuracy of 90. 72\%, 84\%, and 97. 12\% in Office-31, Office-Home and Office-Caltech datasets, respectively. Further evaluations confirm that our study improves existing methods by an average classification accuracy increment of +1.23\%, +7.26\%, and +1.77\% on the respective datasets.

A Source-Free Approach for Domain Adaptation via Multiview Image Transformation and Latent Space Consistency

TL;DR

The paper tackles domain adaptation without source data by introducing a source-free framework that learns domain-invariant features from the target domain through multiview augmentation and latent space consistency. A ConvNeXt-based encoder maps inputs to a latent space, where two augmented views of each target sample are enforced to have similar latent representations, via a combined loss . Key contributions include the latent space consistency mechanism, a concrete two-view augmentation strategy, and empirical validation on Office-31, Office-Home, and Office-Caltech showing strong performance and robustness without adversarial training or pseudo-labeling. The approach offers a practical, efficient alternative for SFDA with potential impact on real-world deployment where access to source data is restricted. Overall, the method demonstrates competitive accuracy gains and consistent domain-invariant feature learning directly from the target domain.

Abstract

Domain adaptation (DA) addresses the challenge of transferring knowledge from a source domain to a target domain where image data distributions may differ. Existing DA methods often require access to source domain data, adversarial training, or complex pseudo-labeling techniques, which are computationally expensive. To address these challenges, this paper introduces a novel source-free domain adaptation method. It is the first approach to use multiview augmentation and latent space consistency techniques to learn domain-invariant features directly from the target domain. Our method eliminates the need for source-target alignment or pseudo-label refinement by learning transferable representations solely from the target domain by enforcing consistency between multiple augmented views in the latent space. Additionally, the method ensures consistency in the learned features by generating multiple augmented views of target domain data and minimizing the distance between their feature representations in the latent space. We also introduce a ConvNeXt-based encoder and design a loss function that combines classification and consistency objectives to drive effective adaptation directly from the target domain. The proposed model achieves an average classification accuracy of 90. 72\%, 84\%, and 97. 12\% in Office-31, Office-Home and Office-Caltech datasets, respectively. Further evaluations confirm that our study improves existing methods by an average classification accuracy increment of +1.23\%, +7.26\%, and +1.77\% on the respective datasets.
Paper Structure (16 sections, 7 equations, 3 figures, 4 tables)

This paper contains 16 sections, 7 equations, 3 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. Methodology
  4. Problem Objective
  5. Initial Training
  6. Latent Space Consistency with Multiview Augmentation
  7. Combined Objective Function for Classification and Consistency
  8. Experimental Setup
  9. Datasets
  10. Model Training
  11. Results
  12. Results Analysis
  13. Ablation Study
  14. t-SNE Visualization
  15. Discussion
  16. ...and 1 more sections

Figures (3)

  • Figure 1: Proposed methodology illustrating the latent feature extraction, classification, multiview representation with augmentations and adaptation with consistency loss for domain adaptation.
  • Figure 2: Sample images from the datasets.
  • Figure 3: t-SNE map showing clustering results with Silhouette scores, Davies-Bouldin Index (DBI), and Calinski-Harabasz Index (CHI) values for different datasets. The comparison between the raw dataset (top row) and the method using ConvNeXt as a backbone (bottom row) highlights improvements in clustering quality as indicated by higher Silhouette scores and lower DBI and CHI values.