Joint Spatial and Layer Attention for Convolutional Networks

TL;DR

This paper introduces a joint spatial and layer attention mechanism for CNNs, enhancing performance in camera pose regression and scene classification by selectively focusing on features and locations.

Contribution

It presents a novel attention approach that dynamically attends to CNN layers and spatial regions, improving accuracy on vision tasks compared to prior methods.

Findings

Reduces median localization error by 18.8% for position.

Improves scene classification accuracy by 3.4%.

Demonstrates effectiveness on multiple benchmark datasets.

Abstract

In this paper, we propose a novel approach that learns to sequentially attend to different Convolutional Neural Networks (CNN) layers (i.e., ``what'' feature abstraction to attend to) and different spatial locations of the selected feature map (i.e., ``where'') to perform the task at hand. Specifically, at each Recurrent Neural Network (RNN) step, both a CNN layer and localized spatial region within it are selected for further processing. We demonstrate the effectiveness of this approach on two computer vision tasks: (i) image-based six degree of freedom camera pose regression and (ii) indoor scene classification. Empirically, we show that combining the ``what'' and ``where'' aspects of attention improves network performance on both tasks. We evaluate our method on standard benchmarks for camera localization (Cambridge, 7-Scenes, and TUM-LSI) and for scene classification (MIT-67 Indoor Scenes). For camera localization our approach reduces the median error by 18.8\% for position and 8.2\% for orientation (averaged over all scenes), and for scene classification it improves the mean accuracy by 3.4\% over previous methods.