PAGen: Phase-guided Amplitude Generation for Domain-adaptive Object Detection

TL;DR

This paper introduces PAGen, a simple frequency domain style adaptation method for domain-adaptive object detection that improves performance without adding inference complexity.

Contribution

We propose a lightweight frequency domain style adaptation approach for UDA in object detection, avoiding complex adversarial training and auxiliary models.

Findings

Significant performance improvements on multiple benchmarks.

No additional computational overhead during inference.

Effective in challenging target domain conditions.

Abstract

Unsupervised domain adaptation (UDA) greatly facilitates the deployment of neural networks across diverse environments. However, most state-of-the-art approaches are overly complex, relying on challenging adversarial training strategies, or on elaborate architectural designs with auxiliary models for feature distillation and pseudo-label generation. In this work, we present a simple yet effective UDA method that learns to adapt image styles in the frequency domain to reduce the discrepancy between source and target domains. The proposed approach introduces only a lightweight pre-processing module during training and entirely discards it at inference time, thus incurring no additional computational overhead. We validate our method on domain-adaptive object detection (DAOD) tasks, where ground-truth annotations are easily accessible in source domains (e.g., normal-weather or synthetic conditions) but challenging to obtain in target domains (e.g., adverse weather or low-light scenes). Extensive experiments demonstrate that our method achieves substantial performance gains on multiple benchmarks, highlighting its practicality and effectiveness.