Strong-Weak Distribution Alignment for Adaptive Object Detection

By Kuniaki Saito, Yoshitaka Ushiku, Tatsuya Harada, Kate Saenko. Computer Vision and Pattern Recognition, 2018

756 citations recorded by the source corpus.

We propose an approach for unsupervised adaptation of object detectors from label-rich to label-poor domains which can significantly reduce annotation costs associated with detection. Recently, approaches that align distributions of source and target images using an adversarial loss have been proven effective for adapting object classifiers. However, for object detection, fully matching the entire distributions of source and target images to each other at the global image level may fail, as domains could have distinct scene layouts and different combinations of objects. On the other hand, strong matching of local features such as texture and color makes sense, as it does not change category level semantics. This motivates us to propose a novel method for detector adaptation based on strong local alignment and weak global alignment. Our key contribution is the weak alignment model, which focuses the adversarial alignment loss on images that are globally similar and puts less emphasis on aligning images that are globally dissimilar. Additionally, we design the strong domain alignment model to only look at local receptive fields of the feature map. We empirically verify the effectiveness of our method on four datasets comprising both large and small domain shifts. Our code is available at https://github.com/VisionLearningGroup/DA_Detection.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/5ca9b18448a2ef8a2443445269ccb151aa100de9

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computer Vision and Pattern Recognition

Citations in the corpus

Exploring Categorical Regularization for Domain Adaptive Object Detection (2020, methodology): “Thanks to the flexibility of Faster R-CNN, recently, it is widely adapted for domain adaptive object detection [1, 28, 41, 12] with adversarial training [5].”
Adapting Object Detectors with Conditional Domain Normalization (2020, methodology): “d 3D ROIs proposal stage (instance alignment). 5 Experimental Results and Analysis 5.1 Results on Cityscapes to Foggy Cityscapes We compare CDN with the state-of-the-art methods in Table 1. Following [34, 47], we also report results using Faster R-CNN model with VGG16 backbone. As shown in Table 1, ”
Semi-Supervised Domain Adaptation via Minimax Entropy (2019, methodology): “UDA has been applied to various applications such as image classification [27], semantic segmentation [32], and object detection [6, 29].”
A Survey of Deep Learning-Based Object Detection (2019, background): “[278] propose a weak alignment model to focus on similarity between different images from domains with large discrepancy rather than aligning images that are globally dissimilar.”
Cross-Domain Adaptive Teacher for Object Detection (2021, methodology): “Following [32, 35], the dataset is combined from PASCAL VOC 2007 and 2012 with total 16,551 images.”
Harmonizing Transferability and Discriminability for Adapting Object Detectors (2020, methodology): “We used the training set during training and evaluated on the testing set by following [44].”
Unbiased Mean Teacher for Cross-domain Object Detection (2020, methodology): “Various approaches have been proposed to address these issues, including instance and image-level adversarial training [6], strong and weak adversarial training [37], graph-based consistency [1], etc.”
Progressive Domain Adaptation for Object Detection (2019, methodology): “Table 3 shows that our method reduces the domain gap across weather conditions and performs favorably against the state-of-the-art methods [3, 25, 37, 16].”
Spatial Attention Pyramid Network for Unsupervised Domain Adaptation (2020, methodology): “For fair comparison, we set the shorter side of the image to 600 following the implementation of [34,36] with RoIAlign [11] in object detection; for instance segmentation and semantic segmentation, we use the same settings as previous methods.”
ST3D: Self-training for Unsupervised Domain Adaptation on 3D Object Detection (2021, methodology): “In contrast to the intensive studies on UDA of the 2D image setting [10, 26, 17, 7, 32, 11, 12], few efforts [41] have been made to explore UDA for 3D detection.”
Back to Reality: Weakly-supervised 3D Object Detection with Shape-guided Label Enhancement (2022, methodology): “Focal loss [17, 34] is utilized to apply weak alignment:”