Abstract:Convolutional neural networks (CNNs) have made remarkable progress in detecting vehicle objects in normal weather conditions. However, the performance of these networks deteriorates when faced with rain and fog, as these conditions degrade image quality and cause blurring. The network models trained on clear images perform poorly on rainy and foggy images due to the differences in distribution between normal weather and adverse weather conditions, leading to domain bias. To address this challenge, we present a novel algorithm called DAGL-Faster (Domain Adaptive Global-Local Alignment Faster RCNN) , which enables domain-adaptive vehicle object detection specifically for rainy and foggy weather. DAGL-Faster extends the Faster RCNN framework by incorporating three domain classifiers. These classifiers aid the network in extracting features that are invariant to the domain differences between the source domain (normal weather) and the target domains (rain or fog). The algorithm tackles the domain dissimilarities from three perspectives: local image-level, global image-level, and instance-level. Additionally, it introduces consistency regularization to facilitate simultaneous alignment at the image-level and instance-level, optimizing the overall alignment effect. Through extensive experiments, we demonstrate the efficacy of DAGL-Faster on two benchmark datasets: Foggy Cityscapes and Rain Vehicle Color-24. The algorithm achieves an impressive mean average precision (mAP) of up to 36.7% on the Foggy Cityscapes dataset and 49.79% on the Rain Vehicle Color-24 dataset. Moreover, DAGL-Faster demonstrates superior computational efficiency, with a processing time of 1.9 seconds per image using a single GTX 1080 Ti GPU. These results surpass state-of-the-art algorithms for popular domain adaptive object detection methods.

Progressive Domain Adaptive Object Detection Based on Self‐Attention in Foggy Weather

Joint Feature-Level And Pixel-Level Domain Adaption For Object Detection In The Wild

Domain Adaptive Object Detection for Autonomous Driving under Foggy Weather

Domain Adaptation based Object Detection for Autonomous Driving in Foggy and Rainy Weather

Object Detection in Foggy Scenes by Embedding Depth and Reconstruction into Domain Adaptation

Progressive Domain Adaptation for Object Detection

Prior-Based Domain Adaptive Object Detection for Hazy and Rainy Conditions

DA-RAW: Domain Adaptive Object Detection for Real-World Adverse Weather Conditions

Multi-Target Domain Adaptation via Unsupervised Domain Classification for Weather Invariant Object Detection

Multi-Level Domain Adaptive Learning For Cross-Domain Detection

Cross-domain Object Detection through Coarse-to-Fine Feature Adaptation

Target Detection Algorithm in Foggy Scenes Based on Dual Subnets

C2FDA: Coarse-to-Fine Domain Adaptation for Traffic Object Detection

A Step-Wise Domain Adaptation Detection Transformer for Object Detection under Poor Visibility Conditions

FogAdapt: Self-Supervised Domain Adaptation for Semantic Segmentation of Foggy Images

Object Detection in Adverse Weather for Autonomous Vehicles Based on Sensor Fusion and Incremental Learning

DAGL-Faster: Domain adaptive faster r-cnn for vehicle object detection in rainy and foggy weather conditions

Spatial Alignment for Unsupervised Domain Adaptive Single-Stage Object Detection

Multilevel Unsupervised Domain Adaptation for Single-Stage Object Detection in Remote Sensing Images

Deep Domain Adaptive Object Detection: a Survey

Plug-and-Play Robust Aerial Object Detection Under Hazy Conditions