Exploring Test-Time Adaptation for Object Detection in Continually Changing Environments

2024-06-24

arXiv.org

AI Summary

Addressing the dual challenges of low-quality pseudo-labels and catastrophic forgetting in Continual Test-Time Adaptation (CTTA) for object detection, this paper proposes an online domain adaptation framework tailored for dynamic object detection. First, a class-adaptive confidence thresholding mechanism is introduced to dynamically generate high-quality, object-level pseudo-labels. Second, a selective random parameter restoration strategy is employed, which resets only non-active parameters to preserve critical knowledge while mitigating forgetting. Third, an object-level contrastive learning module is integrated to enhance cross-domain feature discriminability. Evaluated on four CTTA detection benchmarks, the method consistently outperforms state-of-the-art approaches: it achieves a +3.2 mAP improvement on the Cityscapes-to-Cityscapes-C task and accelerates adaptation efficiency by 20%, demonstrating both effectiveness and robustness under continual test-time distribution shifts.

Technology Category

• Machine Learning: Transfer, Domain Adaptation, Multi-Task Learning
• Computer Vision: Adversarial Attacks & Robustness
• Search and Optimization: Learning to Search

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Economics, Online Markets and Human Computation: Data quality aspects of human-annotated datasets
• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems

Abstract

Real-world application models are commonly deployed in dynamic environments, where the target domain distribution undergoes temporal changes. Continual Test-Time Adaptation (CTTA) has recently emerged as a promising technique to gradually adapt a source-trained model to continually changing target domains. Despite recent advancements in addressing CTTA, two critical issues remain: 1) Fixed thresholds for pseudo-labeling in existing methodologies lead to low-quality pseudo-labels, as model confidence varies across categories and domains; 2) Stochastic parameter restoration methods for mitigating catastrophic forgetting fail to preserve critical information effectively, due to their intrinsic randomness. To tackle these challenges for detection models in CTTA scenarios, we present AMROD, featuring three core components. Firstly, the object-level contrastive learning module extracts object-level features for contrastive learning to refine the feature representation in the target domain. Secondly, the adaptive monitoring module dynamically skips unnecessary adaptation and updates the category-specific threshold based on predicted confidence scores to enable efficiency and improve the quality of pseudo-labels. Lastly, the adaptive randomized restoration mechanism selectively reset inactive parameters with higher possibilities, ensuring the retention of essential knowledge. We demonstrate the effectiveness of AMROD on four CTTA object detection tasks, where AMROD outperforms existing methods, especially achieving a 3.2 mAP improvement and a 20% increase in efficiency on the Cityscapes-to-Cityscapes-C CTTA task. The code will be released.

Problem

• Adapting object detection models to dynamic environments
• Improving pseudo-label quality with adaptive thresholds
• Mitigating catastrophic forgetting via selective parameter restoration

Innovation

• Object-level contrastive learning for feature refinement
• Adaptive monitoring for dynamic pseudo-label thresholding
• Selective parameter restoration to retain critical knowledge