Visual Instance Retrieval with Deep Convolutional Networks

TL;DR

This paper investigates how convolutional neural network features can be optimized for visual instance retrieval, emphasizing multi-scale local features and geometric invariance to outperform existing methods.

Contribution

It introduces an efficient pipeline for extracting local features from ConvNets, incorporating multi-scale and geometric invariance considerations for improved retrieval performance.

Findings

ConvNet features outperform state-of-the-art methods when extracted properly

Multi-scale schemes and geometric invariance improve retrieval accuracy

Experiments on five standard datasets validate the approach

Abstract

This paper provides an extensive study on the availability of image representations based on convolutional networks (ConvNets) for the task of visual instance retrieval. Besides the choice of convolutional layers, we present an efficient pipeline exploiting multi-scale schemes to extract local features, in particular, by taking geometric invariance into explicit account, i.e. positions, scales and spatial consistency. In our experiments using five standard image retrieval datasets, we demonstrate that generic ConvNet image representations can outperform other state-of-the-art methods if they are extracted appropriately.