[Paper Review] Deep Domain Adaptive Object Detection: a Survey

Wanyi Li, Fuyu Li+3 more|arXiv (Cornell University)|Feb 17, 2020

Domain Adaptation and Few-Shot Learning42 references13 citations

TL;DR

This survey reviews deep domain adaptive object detection (DDAOD) methods that address the domain shift problem in object detection by adapting models trained on a source domain to perform well on a target domain with limited or no labeled data. It classifies existing approaches into five categories—domain adaptation in feature space, adversarial training, self-training, knowledge distillation, and meta-learning—and provides a comprehensive analysis of representative methods, offering insights into future research directions in DDAOD.

ABSTRACT

Deep learning (DL) based object detection has achieved great progress. These methods typically assume that large amount of labeled training data is available, and training and test data are drawn from an identical distribution. However, the two assumptions are not always hold in practice. Deep domain adaptive object detection (DDAOD) has emerged as a new learning paradigm to address the above mentioned challenges. This paper aims to review the state-of-the-art progress on deep domain adaptive object detection approaches. Firstly, we introduce briefly the basic concepts of deep domain adaptation. Secondly, the deep domain adaptive detectors are classified into five categories and detailed descriptions of representative methods in each category are provided. Finally, insights for future research trend are presented.

Motivation & Objective

• To address the domain shift problem in deep object detection where training and test data distributions differ.
• To review state-of-the-art methods in deep domain adaptive object detection (DDAOD) under real-world data distribution shifts.
• To classify and analyze DDAOD approaches into five distinct categories based on their adaptation mechanisms.
• To provide insights into open challenges and future research trends in DDAOD.

Proposed method

• Classifying DDAOD methods into five categories: domain adaptation in feature space, adversarial training, self-training, knowledge distillation, and meta-learning.
• Analyzing representative methods in each category, including their core architectures and adaptation strategies.
• Evaluating techniques that align feature distributions between source and target domains using domain-invariant representations.
• Surveying methods that leverage unlabeled target data through consistency regularization or pseudo-labeling.
• Examining adversarial frameworks that use discriminators to distinguish source from target features.
• Reviewing knowledge distillation-based approaches that transfer knowledge from a teacher model to a student model trained on the target domain.

Experimental results

Research questions

• RQ1 How do different DDAOD methods mitigate domain shift in object detection?
• RQ2 What are the key architectural and training strategies used in state-of-the-art DDAOD approaches?
• RQ3 How do self-supervised and adversarial methods compare in adapting object detectors across domains?
• RQ4 What are the limitations of current DDAOD methods in handling large domain shifts or long-tail distributions?
• RQ5 What future research directions are most promising for advancing DDAOD?

Key findings

• The survey identifies five main categories of DDAOD methods, each with distinct strengths in handling domain shift.
• Adversarial training and self-training methods show strong performance in aligning feature distributions across domains.
• Knowledge distillation-based approaches effectively transfer detection knowledge from source to target domains with minimal annotations.
• Meta-learning methods demonstrate potential in few-shot domain adaptation scenarios.
• The survey highlights the need for more robust, generalizable, and scalable DDAOD frameworks for real-world deployment.
• Future research should focus on improving generalization under large domain shifts, handling long-tail distributions, and integrating uncertainty estimation.