Domain Generalization for Object Recognition with Multi-task Autoencoders

TL;DR

This paper introduces Multi-Task Autoencoders (MTAE), a novel feature learning method that enhances cross-domain object recognition by learning domain-invariant features through transforming images into multiple related domains.

Contribution

The paper proposes MTAE, extending denoising autoencoders to utilize natural inter-domain variability for improved domain generalization in object recognition.

Findings

MTAE outperforms existing autoencoder-based models.

MTAE surpasses state-of-the-art algorithms in domain generalization.

Features learned by MTAE are robust across unseen domains.

Abstract

The problem of domain generalization is to take knowledge acquired from a number of related domains where training data is available, and to then successfully apply it to previously unseen domains. We propose a new feature learning algorithm, Multi-Task Autoencoder (MTAE), that provides good generalization performance for cross-domain object recognition. Our algorithm extends the standard denoising autoencoder framework by substituting artificially induced corruption with naturally occurring inter-domain variability in the appearance of objects. Instead of reconstructing images from noisy versions, MTAE learns to transform the original image into analogs in multiple related domains. It thereby learns features that are robust to variations across domains. The learnt features are then used as inputs to a classifier. We evaluated the performance of the algorithm on benchmark image recognition datasets, where the task is to learn features from multiple datasets and to then predict the image label from unseen datasets. We found that (denoising) MTAE outperforms alternative autoencoder-based models as well as the current state-of-the-art algorithms for domain generalization.