Multi-representation adaptation network for cross-domain image classification

Abstract

Image processing is a method to perform some operations on an image, in order to get an enhanced image or to extract some useful information from it. The conventional image processing algorithms cannot perform well in scenarios where the training images (source domain) that are used to learn the model have a different distribution with test images (target domain). Also, many real world applications suffer from a limited number of training labeled data and therefore benefit from the related available labeled datasets to train the model. But manual labeling of sufficient training data for diverse application domains is a costly, laborious task and often prohibitive. Therefore, designing models that can leverage rich labeled data in one domain and be applicable to a different but related domain is highly desirable. In particular, domain adaptation or transfer learning algorithms seek to generalize a model trained in a source domain to a new target domain. Recent years has witnessed increasing interest in these types of models due to their practical importance in real-life applications. Existing approaches largely balance the distributions of representations extracted by a single structure, and representations may contain only partial information, for example, only part of the color, brightness, and saturation information. Therefore, we propose multi-representation adaptation, which can dramatically improve the classification accuracy for image classification, and in particular aims to align the multi-representation distributions extracted by a hybrid structure called the Adaptation Module. Accordingly, we provide multi-representation adaption to perform the task of image classification through multi-representation alignment that can capture information from different angles. In addition, we use the Maximum Mean Difference to calculate the matching loss. Our approach can be easily implemented by further expanding feed-in models with Adaptation Module, and the network can be effectively trained through savings diffusion. Experiments performed on two standard image datasets show the effectiveness of suggested method.

Key Words : Domain adaptation, Multi-representation, Maximum mean discrepancy, Adaptation module, Conditional distribution