Disentangled Geometric Alignment with Adaptive Contrastive Perturbation for Reliable Domain Transfer

TL;DR

GAMA++ introduces a disentangled geometric alignment framework with adaptive contrastive perturbation and cross-domain consistency, significantly improving domain transfer performance by better aligning semantic structures and handling class-specific variations.

Contribution

It presents GAMA++, a novel approach that disentangles manifold factors and adapts perturbations for enhanced domain transfer, outperforming existing methods.

Findings

Achieves state-of-the-art results on multiple benchmarks.

Improves class-level alignment fidelity.

Enhances boundary robustness in domain adaptation.

Abstract

Despite progress in geometry-aware domain adaptation, current methods such as GAMA still suffer from two unresolved issues: (1) insufficient disentanglement of task-relevant and task-irrelevant manifold dimensions, and (2) rigid perturbation schemes that ignore per-class alignment asymmetries. To address this, we propose GAMA++, a novel framework that introduces (i) latent space disentanglement to isolate label-consistent manifold directions from nuisance factors, and (ii) an adaptive contrastive perturbation strategy that tailors both on- and off-manifold exploration to class-specific manifold curvature and alignment discrepancy. We further propose a cross-domain contrastive consistency loss that encourages local semantic clusters to align while preserving intra-domain diversity. Our method achieves state-of-the-art results on DomainNet, Office-Home, and VisDA benchmarks under both standard and few-shot settings, with notable improvements in class-level alignment fidelity and boundary robustness. GAMA++ sets a new standard for semantic geometry alignment in transfer learning.