Learning Attributes Equals Multi-Source Domain Generalization

By Chuang Gan, Tianbao Yang, Boqing Gong. 2016

206 citations recorded by the source corpus.

Attributes possess appealing properties and benefit many computer vision problems, such as object recognition, learning with humans in the loop, and image retrieval. Whereas the existing work mainly pursues utilizing attributes for various computer vision problems, we contend that the most basic problem-how to accurately and robustly detect attributes from images-has been left under explored. Especially, the existing work rarely explicitly tackles the need that attribute detectors should generalize well across different categories, including those previously unseen. Noting that this is analogous to the objective of multi-source domain generalization, if we treat each category as a domain, we provide a novel perspective to attribute detection and propose to gear the techniques in multi-source domain generalization for the purpose of learning cross-category generalizable attribute detectors. We validate our understanding and approach with extensive experiments on four challenging datasets and three different problems.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Radiology, Nuclear Medicine and Imaging

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

In Radiology, Nuclear Medicine and Imaging

Citations in the corpus

Semantic Compositional Networks for Visual Captioning (2016, methodology): “Specifically, [18] applied retrieved sentences as additional semantic information to guide the LSTM when generating captions, while [13, 49, 54] applied a semantic-concept-detection process [15] before generating sentences.”
Domain Adaptation for Visual Applications: A Comprehensive Survey (2017, methodology): “is to explore the unlabeled source data and to discover repetitive patterns that could be used for the supervised learning task. On the other hand, DA is more closely related to domain generalization [221, 222, 138, 223, 224], multitask learning [225, 226, 227] or few-shot learning [228, 229] discus”
A Comprehensive Survey of Deep Learning for Image Captioning (2018, methodology): “However, Wu et al. Recent semantic concept based image captioning methods [150, 156] applied semantic-concept-detection process [40]to obtain explicit semantic concepts.”
FSDR: Frequency Space Domain Randomization for Domain Generalization (2021, background): “invariant features without requiring target-domain data in training [77, 44, 11, 36, 33, 34].”
Domain Randomization and Pyramid Consistency: Simulation-to-Real Generalization Without Accessing Target Domain Data (2019, background): “Domain Generalization In contrast to Domain Adaptation, where the network is tested on a known target domain, and the images in the target domain, although without labels, are accessible during the training process, Domain Generalization is tested on unseen domains [31, 12].”
Zero-Shot Learning with Generative Latent Prototype Model (2017, background): “The authors are with the College of Computer Science, Zhejiang University, Hangzhou, 310027, China e-mail: (ynli@zju.edu.cn, dhwang@zju.edu.cn). rewarding [16].”
Test-Time Training with Self-Supervision for Generalization under Distribution Shifts (2019, background): “…multiple environment distributions, some of which are available during training (source), while others are used for testing (target) (Li et al., 2018; Shankar et al., 2018; Muandet et al., 2013; Balaji et al., 2018; Ghifary et al., 2015; Motiian et al., 2017; Li et al., 2017a; Gan et al., 2016).”
Domain Generalization: A Survey (2021, methodology): “In [29], pseudo-labels are assigned to unlabeled source data and an off-the-shelf style transfer model is used to augment the domain space.”
Spatial-Aware Object Embeddings for Zero-Shot Localization and Classification of Actions (2017, methodology): “Inspired by zero-shot image classification [24], several works have performed zero-shot action classification by learning a mapping of actions to attributes [11, 30, 64].”
Learning to Generate Novel Domains for Domain Generalization (2020, unknown): “Many DG methods are based on the idea of domain alignment popularized from the UDA literature [12], with a goal to learn a domain-invariant representation by minimizing the domain discrepancy between sources [36,15,11,16,29,28].”
I Have Seen Enough: Transferring Parts Across Categories (2016, background): “This is a fundamental difference from our case, where no target samples are available, also known as domain generalization [14, 16, 34] or predictive domain adaptation [62].”
Towards Out-Of-Distribution Generalization: A Survey (2021, unknown): “…[48], [49] [37], [39], [50], [51], [51], [52], [53], [53], [54], [54], [55], [56], [57], [58], [59], [60] [6], [61], [62], [63], [63], [64], [65], [66], [67], [68], [69], [70](Section 4.1.1) [52], [71], [72], [73], [74], [75], [76], [77], [78], [79], [80], [81], [82] [83], [84], [85], [86], [87],…”