Hierarchical contrastive adaptation for cross-domain object detection

Deng, Ziwei; Kong, Qing; Akira, Naoto; Yoshinaga, Tomoaki

doi:10.1007/s00138-022-01317-7

Cited by 7 publications

(3 citation statements)

References 45 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Hong et al [21] handle these issues as a dictionary learning problem, where the spectral variability dictionary and estimation of the abundance maps are learned simultaneously. For a labeled source dataset and an unlabeled target dataset, unsupervised domain adaptation methods generalize the model by aligning source and target [36]. Cheng adjusts the decision boundary biased towards the target data source domain and adds adversarial training in conjunction with image-to-image translation techniques [37].…”

Section: Related Workmentioning

confidence: 99%

Unsupervised Domain Adaptation With Debiased Contrastive Learning and Support-Set Guided Pseudolabeling for Remote Sensing Images

Biswas,

Tešić

2024

IEEE J. Sel. Top. Appl. Earth Observations Remote Sensing

View full text Add to dashboard Cite

The variability in different altitudes, geographical variances, and weather conditions across datasets degrade stateof-the-art (SOTA) DNN object detection performance. Unsupervised and Semi-supervised domain adaptation (DA) have been decent solutions to bridge the gap between two different distributions of datasets. The state-of-the-art pseudo-labeling process is susceptible to background noise, hindering the optimal performance in target datasets. The existing contrastive DA methods overlook the bias effect introduced from the false negative (FN) target samples, which misleads the complete learning process. This paper proposes DCLDA (support-guided debiased contrastive learning for domain adaptation) to properly label the unlabeled target dataset and remove the bias toward target detection. We introduce (i) A support-set curated approach to generate high-quality pseudo-labels from the target dataset proposals, (ii) a reduced distribution gap across different datasets using domain alignment on local, global, and instance-aware features for remote sensing datasets, and (iii) novel debiased contrastive loss function, that makes the model more robust for the variable appearance of a particular class over images and domains. The proposed debiased contrastive learning pivots on class probabilities to address the challenge of false negatives in the unsupervised framework. Our model outperforms the compared SOTA models with a minimum gain of +3.9%, +3.2%, +12.7%, and +2.1% of mAP for DIOR, DOTA, Visdrone, and UAVDT datasets, respectively.

show abstract

Section: Related Workmentioning

confidence: 99%

Unsupervised Domain Adaptation With Debiased Contrastive Learning and Support-Set Guided Pseudolabeling for Remote Sensing Images

Biswas,

Tešić

2024

IEEE J. Sel. Top. Appl. Earth Observations Remote Sensing

View full text Add to dashboard Cite

show abstract

“…Domain adaptation has risen as one of the approaches to speed up the pseudo-labeling of objects in the target domain using source labels. Domain adaptation for object recognition in consumer image datasets successfully addresses weather, lighting conditions, geological variance, variation in image quality, and cross-camera adaptation by aligning the global feature distribution of data from the origin and target domains [15]. Recent State-of-the-art (SOTA) work of unsupervised domain adaptation for aerial imagery uses the reconstructed feature alignment method instead of adversarialbased feature alignment to avoid background noise alignment [16].…”

Section: Domain Adaptation With Contrastive Learning For Object Detec...mentioning

confidence: 99%

“…Feature representation of the source and target objects differs in the feature space, and there is a huge gap due to lightning, geographic, weather, and acquisition differences, and the difference is illustrated by a green line in Figure 7. Contrastive learning brings similar points to close together and pushes dissimilar points separate from each other by calculating similarities between pairs [42], [15]. A pair of feature vectors with high similarity are placed close together, and vector pairs with low similarity are placed distantly in feature space.…”

Section: B Lgda Model: Local Global Domain Adaptation Modelmentioning

confidence: 99%

Domain Adaptation with Contrastive Learning for Object Detection in Satellite Imagery

Biswas,

Tesic

2023

Preprint

View full text Add to dashboard Cite

<p>State-of-the-art object detection methods applied to satellite and drone imagery largely fail to identify small and dense objects. One reason is the high variability of content in the overhead imagery due to the terrestrial region captured and the high variability of acquisition conditions. Another reason is that the number and size of objects in aerial imagery are very different than in the consumer data. In this work, we propose a small object detection pipeline that improves the feature extraction process by spatial pyramid pooling, cross-stage partial networks, heatmap-based region proposal network, and object localization and identification through a novel image difficulty score that adapts the overall focal loss measure based on the image difficulty. Next, we propose novel contrastive learning with progressive domain adaptation to produce domain-invariant features across aerial datasets using local and global components. We show we can alleviate the degradation of object identification in previously unseen datasets. We create a first-ever domain adaptation benchmark using contrastive learning for the object detection task in highly imbalanced satellite datasets with significant domain gaps and dominant small objects from existing satellite benchmarks—the proposed method results in up to a 7.4% increase in mAP performance measure over the best state-of-art. </p>

show abstract

Domain disentanglement and contrastive learning with source-guided sampling for unsupervised domain adaptation person re-identification

Wu,

Liu,

Chen

et al. 2024

Machine Vision and Applications

View full text Add to dashboard Cite

Hierarchical contrastive adaptation for cross-domain object detection

Cited by 7 publications

References 45 publications

Unsupervised Domain Adaptation With Debiased Contrastive Learning and Support-Set Guided Pseudolabeling for Remote Sensing Images

Unsupervised Domain Adaptation With Debiased Contrastive Learning and Support-Set Guided Pseudolabeling for Remote Sensing Images

Domain Adaptation with Contrastive Learning for Object Detection in Satellite Imagery

Domain disentanglement and contrastive learning with source-guided sampling for unsupervised domain adaptation person re-identification

Contact Info

Product

Resources

About