Collaborative Training Between Region Proposal Localization and Classification for Domain Adaptive Object Detection

Zhao, Ganlong; Li, Guanbin; Xu, Ruijia; Li, Lin

doi:10.1007/978-3-030-58523-5_6

Cited by 66 publications

(36 citation statements)

References 26 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Note that the oracle performance of training Faster R-CNN with target domain labels is only 43.52% and there remains little room for further improvement. [22] and CT [54], our method outperforms them by +7.26% and +5.31% respectively, demonstrating the importance of our uncertainty-guided self-training. Our full model also boosts the performance by +3.7% over our baseline model.…”

Section: 82mentioning

confidence: 80%

“…Faster R-CNN [36] (Source) 34.60 DA-Faster (CVPR'18) [5] 41.90 Noisy Labeling (ICCV'19) [22] 42.56 SWDA (CVPR'19) [39] 47.70 GPA (CVPR'20) [51] 47.60 CT (ECCV'20) [54] 44.51 MeGA-CDA (CVPR'21) [45] 44.80…”

Section: Methods Ap (Car)mentioning

confidence: 99%

“…The reconstruction-based methods [19,37] utilize image style translation techniques, such as CycleGAN [55], to diminish the gap between source and target domain. Recently, selftraining based methods [38,22,54] have been proposed as a simpler yet effective strategy, which explores the pseudolabels from the target domain to re-train model.…”

Section: Cross-domain Object Detectionmentioning

confidence: 99%

“…13) and full model (Eq. 14) with Faster R-CNN [36] and recent state-of-theart methods: 1) Adversarial feature alignment based meth-ods including DA-Faster [5], SWDA [39], ICR-CCR [50], GPA [51], and MeGA-CDA [45]; 2) Self-training based methods including Noisy Labeling [22], and CT [54].…”

Section: Performance Comparisonmentioning

confidence: 99%

See 3 more Smart Citations

Uncertainty-Aware Model Adaptation for Unsupervised Cross-Domain Object Detection

Cai,

Luo,

Zhong

et al. 2021

Preprint

View full text Add to dashboard Cite

This work tackles the unsupervised cross-domain object detection problem which aims to generalize a pre-trained object detector to a new target domain without labels. We propose an uncertainty-aware model adaptation method for object detection, which is based on two motivations: 1) the estimation and exploitation of model uncertainty in a new domain is critical for reliable domain adaptation; and 2) the joint alignment of distributions for inputs (feature alignment) and outputs (model self-training) is needed. To this end, we compose a Bayesian CNN-based framework for uncertainty estimation in object detection, and propose an algorithm for generation of uncertainty-aware pseudolabels. We also devise a scheme for joint feature alignment and self-training of the object detection model with pseudolabels. Experiments on multiple cross-domain object detection benchmarks show that our proposed method achieves state-of-the-art performance.

show abstract

Section: 82mentioning

confidence: 80%

Section: Methods Ap (Car)mentioning

confidence: 99%

Section: Cross-domain Object Detectionmentioning

confidence: 99%

Section: Performance Comparisonmentioning

confidence: 99%

See 2 more Smart Citations

Uncertainty-Aware Model Adaptation for Unsupervised Cross-Domain Object Detection

Cai,

Luo,

Zhong

et al. 2021

Preprint

View full text Add to dashboard Cite

show abstract

“…Generalization under domain shifts is a core problem in machine learning and computer vision. Many [3,17,51,47,46,50,38,1,21,33] have designed domain adaptive object detectors, which leverage the unlabeled target domain data through feature alignment. Sun et al [40] uses an auxiliary rotation task to leverage the unlabeled target domain data for semantic segmentation.…”

Section: Related Workmentioning

confidence: 99%

Robust Object Detection via Instance-Level Temporal Cycle Confusion

Wang

Huang

Liu

et al. 2021

Preprint

View full text Add to dashboard Cite

Building reliable object detectors that are robust to domain shifts, such as various changes in context, viewpoint, and object appearances, is critical for real-world applications. In this work, we study the effectiveness of auxiliary self-supervised tasks to improve out-of-distribution generalization of object detectors. Inspired by the principle of maximum entropy, we introduce a novel self-supervised task, instance-level temporal cycle confusion (CycConf), which operates on the region features of the object detectors. For each object, the task is to find the most different object proposals in the adjacent frame in a video and then cycle back to itself for self-supervision. CycConf encourages the object detector to explore invariant structures across instances under various motions, which leads to improved model robustness in unseen domains at test time. We observe consistent outof-domain performance improvements when training object detectors in tandem with self-supervised tasks on large-scale video datasets (BDD100K and Waymo open data). The joint training framework also establishes a new state-of-the-art on standard unsupervised domain adaptative detection benchmarks (Cityscapes, Foggy Cityscapes, and Sim10K). 1 * Equal contribution. The authors are listed in alphabetical order.

show abstract