AFAN: Augmented Feature Alignment Network for Cross-Domain Object Detection

TL;DR

This paper introduces AFAN, a novel framework for unsupervised cross-domain object detection that combines intermediate domain image generation, multi-scale feature alignment, and instance-level domain invariance to improve performance.

Contribution

AFAN integrates intermediate domain image generation with domain-adversarial training and multi-level feature alignment in a unified model for better domain-invariant object detection.

Findings

Outperforms state-of-the-art on standard benchmarks

Effectively bridges domain divergence with synthetic intermediate images

Learned domain-invariant features across multiple semantic levels

Abstract

Unsupervised domain adaptation for object detection is a challenging problem with many real-world applications. Unfortunately, it has received much less attention than supervised object detection. Models that try to address this task tend to suffer from a shortage of annotated training samples. Moreover, existing methods of feature alignments are not sufficient to learn domain-invariant representations. To address these limitations, we propose a novel augmented feature alignment network (AFAN) which integrates intermediate domain image generation and domain-adversarial training into a unified framework. An intermediate domain image generator is proposed to enhance feature alignments by domain-adversarial training with automatically generated soft domain labels. The synthetic intermediate domain images progressively bridge the domain divergence and augment the annotated source domain training data. A feature pyramid alignment is designed and the corresponding feature discriminator is used to align multi-scale convolutional features of different semantic levels. Last but not least, we introduce a region feature alignment and an instance discriminator to learn domain-invariant features for object proposals. Our approach significantly outperforms the state-of-the-art methods on standard benchmarks for both similar and dissimilar domain adaptations. Further extensive experiments verify the effectiveness of each component and demonstrate that the proposed network can learn domain-invariant representations.