| 123 | 2 | 713 |
| 下载次数 | 被引频次 | 阅读次数 |
夜间目标检测任务中，目标能见度低，难以对图像数据进行大量标注，导致使用监督方法进行大规模数据训练的困难，而当使用较小规模标注数据训练时监督方法容易产生过拟合，预测准确性差。针对这些问题，本文提出一种无监督域适应夜间目标检测模型，使用有标注白昼图像和无标注夜间图像训练。模型使用了昼夜图像增强方法，减小昼夜域间隙并提升夜间训练数据的复杂性以丰富特征学习；将多尺度通道注意力引入Faster-RCNN模型，提升感知多尺度特征的能力；使用类别对比学习方法获得具有鉴别性和域间不变性的类别特征。在城市交通数据集BDD100K和SODA10M上进行的实验表明，本文方法性能优于常用的域适应目标检测方法。
Abstract：In the task of nighttime object detection, the visibility of the object is low, and it is difficult to annotate a large amount of image data, which makes it difficult to use supervised methods for large-scale data training. And when using a small-scale annotated data for training supervised methods tend to overfit on the training data, resulting in poor prediction accuracy. To address these issues, an unsupervised domain adaptation nighttime object detection model using labeled daytime images and unlabeled nighttime images for training is proposed in this paper. Day-night image enhancement methods is employed in the model to reduce the domain gap and enhance the complexity of nighttime training data to enrich feature learning; multi-scale channel attention is introduced to the Faster-RCNN model to enhance its ability to perceive multi-scale features; and class-level contrastive learning is used to obtain discriminative and domain-invariant class features. Experiments conducted on the urban traffic datasets BDD100K and SODA10M demonstrate that the performance of the proposed method exceeds that of commonly used domain adaptation target detection methods.
[1] Cui Z,Li K,Gu L,et al.You only need 90k parameters to adapt light:a light weight transformer for image enhancement and exposure correction[EB/OL].arxiv preprint arxiv:2205.14871.(2022-10-08).https://arxiv.org/pdf/2205.14871.
[2] Liu W,Ren G,Yu R,et al.Image-adaptive YOLO for object detection in adverse weather conditions[C]//Proceedings of the AAAI conference on artificial intelligence.2022,36(2):1792-1800.
[3] Yin X,Yu Z,Fei Z,et al.PE-YOLO:Pyramid enhancement network for dark object detection[C]//International conference on artificial neural networks.Cham:Springer Nature Switzerland,2023:163-174.
[4] Xu X,Wang R,Lu J.Low-Light image enhancement via structure modeling and guidance[C]//Proceedings of the IEEE/CVF conference on computer vision and pattern recognition.2023:9893-9903.
[5] Chen Y,Li W,Sakaridis C,et al.Domain adaptive faster r-cnn for object detection in the wild[C]//Proceedings of the IEEE conference on computer vision and pattern recognition.2018:3339-3348.
[6] He M,Wang Y,Wu J,et al.Cross domain object detection by target-perceived dual branch distillation[C]//Proceedings of the IEEE/CVF conference on computer vision and pattern recognition.2022:9570-9580.
[7] Deng J,Li W,Chen Y,et al.Unbiased mean teacher for cross-domain object detection[C]//Proceedings of the IEEE/CVF Conference on computer vision and pattern recognition.2021:4091-4101.
[8] Li Y J,Dai X,Ma C Y,et al.Cross-domain adaptive teacher for object detection[C]//Proceedings of the IEEE/CVF Conference on computer vision and pattern recognition.2022:7581-7590.
[9] Ren S,He K,Girshick R,et al.Faster r-cnn:Towards real-time object detection with region proposal networks[C]//Advances in neural information processing systems,2015:91-99.
[10] He K,Fan H,Wu Y,et al.Momentum contrast for unsupervised visual representation learning[C]//Proceedings of the IEEE/CVF conference on computer vision and pattern recognition.2020:9729-9738.
[11] Huang J,Guan D,Xiao A,et al.Category contrast for unsupervised domain adaptation in visual tasks[C]//Proceedings of the IEEE/CVF conference on computer vision and pattern recognition.2022:1203-1214.
[12] Yu F,Chen H,Wang X,et al.Bdd100k:A diverse driving dataset for heterogeneous multitask learning[C]//Proceedings of the IEEE/CVF conference on computer vision and pattern recognition.2020:2636-2645.
[13] Han J,Liang X,Xu H,et al.SODA10M:a large-scale 2D self/semi-supervised object detection dataset for autonomous driving[EB/OL].arxiv preprint arxiv:2106.11118.(2021-11-08).https://arxiv.org/abs/2106.11118.
[14] Tarvainen A,Valpola H.Mean teachers are better role models:Weight-averaged consistency targets improve semi-supervised deep learning results[J].Advances in neural information processing systems,2017,30:1195-1204.
[15] Hu J,Shen L,Sun G.Squeeze-and-excitation networks[C]//Proceedings of the IEEE conference on computer vision and pattern recognition.2018:7132-7141.
基本信息:
DOI：10.20064/j.cnki.2095-347X.2024.04.003
中图分类号:TP391.41
引用信息:
[1]卫禹帆,张丽红.基于域适应与类别对比的夜间目标检测方法[J].网络新媒体技术,2024,13(04):16-25.DOI:10.20064/j.cnki.2095-347X.2024.04.003.
基金信息:
山西省研究生创新项目(编号:2023SJ012); 山西省高等学校教学改革创新项目(编号:J2021086)
2023-12-19
2023
2024-01-15
2024-01-15
2024
1
2024-07-15
2024-07-15