Ensemble Diffusion for Retrieval

As a postprocessing procedure, diffusion process has demonstrated its ability of substantially improving the performance of various visual retrieval systems. Whereas, great efforts are also devoted to similarity (or metric) fusion, seeing that only one individual type of similarity cannot fully reveal the intrinsic relationship between objects. This stimulates a great research interest of considering similarity fusion in the framework of diffusion process (i.e., fusion with diffusion) for robust retrieval. In this paper, we firstly revisit representative methods about fusion with diffusion, and provide new insights which are ignored by previous researchers. Then, observing that existing algorithms are susceptible to noisy similarities, the proposed Regularized Ensemble Diffusion (RED) is bundled with an automatic weight learning paradigm, so that the negative impacts of noisy similarities are suppressed. At last, we integrate several recently-proposed similarities with the proposed framework. The experimental results suggest that we can achieve new state-of-the-art performances on various retrieval tasks, including 3D shape retrieval on ModelNet dataset, and image retrieval on Holidays and Ukbench dataset.

1. Introduction

Object retrieval is a fundamental yet hot topic in computer vision, which has attracted much attention for decades. Given a query instance, its target is to find objects sharing similar visual appearances with the query in a large database. For a long time, it is crucial to design discriminative representations, so that the metric defined on the representations can be robust to common deformations, such as rotation, occlusion, illumination, etc. Conventionally, Bag-of-Words (BoW) is usually employed thanks to the design of local descriptors (e.g., [27] for images, [26], [37] for shapes, [42], [43], [15], [14] for 3D models). In recent years, the rapid development of deep learning algorithms and G-PU computing platforms has shifted the research attention to deep-learned features [2], [31], [11], [47], [24], which yield a remarkable performance boost against conventional handcrafted features.