Visual Recognition by Learning From Web Data via Weakly Supervised Domain Generalization

Abstract

In this paper, a weakly supervised domain generalization (WSDG) method is proposed for real-world visual recognition tasks, in which we train classifiers by using Web data (e.g., Web images and Web videos) with noisy labels. In particular, two challenging problems need to be solved when learning robust classifiers, in which the first issue is to cope with the label noise of training Web data from the source domain, while the second issue is to enhance the generalization capability of learned classifiers to an arbitrary target domain. In order to handle the first problem, the training samples within each category are partitioned into clusters, where we use one bag to denote each cluster and instances to denote the samples in each cluster. Then, we identify a proportion of good training samples in each bag and train robust classifiers by using the good training samples, which leads to a multi-instance learning (MIL) problem. In order to handle the second problem, we assume that the training samples possibly form a set of hidden domains, with each hidden domain associated with a distinctive data distribution. Then, for each category and each hidden latent domain, we propose to learn one classifier by extending our MIL formulation, which leads to our WSDG approach. In the testing stage, our approach can obtain better generalization capability by effectively integrating multiple classifiers from different latent domains in each category. Moreover, our WSDG approach is further extended to utilize additional textual descriptions associated with Web data as privileged information (PI), although testing data do not have such PI. Extensive experiments on three benchmark data sets indicate that our newly proposed methods are effective for real-world visual recognition tasks by learning from Web data.

Learning from Noisy Web Data with Category-Level Supervision

“…Webly Supervised Image Classification: In recent years, a great many research works [20], [3], [21], [22], [23], [24], [25] were proposed to handle the label noise when learning from web data, by utilizing weakly supervised learning techniques. More recently, several deep learning approaches were developed for webly supervised learning [4], [1], [2], [5], [6], [26] by using label flip layer, bootstrapping, query expansion, attention model, or multi-instance learning techniques.…”

Fine-grained Classification using Heterogeneous Web Data and Auxiliary Categories

“…Existing domain adaptation approaches can be roughly grouped into reweight based method [33], subspace based method [34], [35], [36], [37], [38], and generative model based method [39], [40]. Recently, many domain generalization methods [41], [42], [43], [44] have been proposed. Different from domain adaptation, domain generalization focuses on the setting in which the unlabeled test instances are unavailable in the training stage.…”

Heuristics of Machine Learning for Home Intrusion Detection Application

“…It is known as artificial intelligence subset. ML algorithms [1] create an example information dependent mathematical method, known as training data, to make assessments or decisions without the functions being specifically designed. Machine learning algorithms are useful in a huge range of applications, such as computer vision and email sorting where designing a basic algorithm is impossible or unfeasible to execute the task effectively.…”