Unsupervised Domain Adaptation Dictionary Learning for Visual Recognition

Abstract

Over the last years, dictionary learning method has been extensively applied to deal with various computer vision recognition applications, and produced state-of-the-art results. However, when the data instances of a target domain have a different distribution than that of a source domain, the dictionary learning method may fail to perform well. In this paper, we address the cross-domain visual recognition problem and propose a simple but effective unsupervised domain adaption approach, where labeled data are only from source domain. In order to bring the original data in source and target domain into the same distribution, the proposed method forcing nearest coupled data between source and target domain to have identical sparse representations while jointly learning dictionaries for each domain, where the learned dictionaries can reconstruct original data in source and target domain respectively. So that sparse representations of original data can be used to perform visual recognition tasks. We demonstrate the effectiveness of our approach on standard datasets. Our method performs on par or better than competitive state-ofthe-art methods.

Few-Shot Object Detection in Unseen Domains

“…Recent domain adaptation methods via label-scarce or none target domain data can be summarized into three main approaches: unsupervised domain adaptation (UDA), domain randomization (DR), and zero-shot domain adaptation (ZDA). UDA [12] strives to transfer knowledge across various domains by mitigating the distributional variations. Commonly, UDA methods attempt either a feature-level adaptation in latent space or a pixel-level adaptation via image-toimage translation techniques.…”

“…A detailed summary of the utilized splits is presented in Table I. Moreover, we split the 20 scenes of real images into 8 training scenes, (2,3,5,6,7,9,11,12), featuring mainly the base classes, and the remaining 12 containing novel training and testing scenes. The inference is done on unseen classes featuring the 11 novel objects.…”

“…Unsupervised domain adaptation (UDA) [11], [12] is a closely related learning paradigm that transfers knowledge from abundant labeled source domain, D S = {X S , Y S }, to a label-scarce target domain, D T = {X T }. X and Y denote the input data samples and the labels, respectively.…”

Unsupervised Video Domain Adaptation: A Disentanglement Perspective

“…Over the past decades, unsupervised domain adaptation (UDA) has attracted extensive research attention (Wilson and Cook 2020). Numerous UDA methods have been proposed and successfully applied to various real-world applications, e.g., object recognition (Tzeng et al 2017;Xiao and Zhang 2021;Zhang et al 2022), semantic segmentation (Zou et al 2018;Kong et al 2021;Saporta et al 2022), and object detection (Cai et al 2019a;Guan et al 2021;Yu et al 2022). However, most of these methods and their applications are limited to the image domain, while much less attention has been devoted to video-based UDA, where the latter is undoubtedly more challenging.…”