Compressive Self-localization Using Relative Attribute Embedding

Abstract-- The use of relative attribute (e.g., beautiful, safe, convenient) -based image embeddings in visual place recognition, as a domain-adaptive compact image descriptor that is orthogonal to the typical approach of absolute attribute (e.g., color, shape, texture) -based image embeddings, is explored in this paper. Most current state-of-the-art visual place recognition (VPR) algorithms employ absolute attribute (e.g., color, shape, texture) -based image embedding for image feature with a boundary condition description [1]-[3] and image similarity search [4]. In this study, we are interested in relative attributes (e.g., beautiful, The objective is to search for the image most relevant to a given query image over an image database. B. Descriptor Similarity The database is constructed as a collection of viewpointannotated Next, descriptor similarity is evaluated between the input view images from visual experiences in the training descriptor R and each database descriptor R Specifically, the procedure for construction consists of two The first method, called binary relative strength (BRS), treats steps (Figure 1): (1) extracting a feature descriptor from the the descriptor as a binary relative attribute (stronger or image, and (2) evaluating the descriptor similarity between weaker), and evaluates the similarity by the query and each database images. Either step is detailed in the following.