Segmentation-Aware Convolutional Networks Using Local Attention Masks

TL;DR

This paper presents a segmentation-aware CNN that integrates local segmentation cues via attention masks, improving spatial precision and performance in dense prediction tasks like semantic segmentation and optical flow.

Contribution

The authors introduce a novel method to incorporate segmentation information into CNNs using local attention masks, enhancing spatial accuracy and performance.

Findings

Achieves performance comparable to DenseCRFs in semantic segmentation.

Produces sharper optical flow responses than baseline networks.

Systematic improvements over strong baseline models in both tasks.

Abstract

We introduce an approach to integrate segmentation information within a convolutional neural network (CNN). This counter-acts the tendency of CNNs to smooth information across regions and increases their spatial precision. To obtain segmentation information, we set up a CNN to provide an embedding space where region co-membership can be estimated based on Euclidean distance. We use these embeddings to compute a local attention mask relative to every neuron position. We incorporate such masks in CNNs and replace the convolution operation with a "segmentation-aware" variant that allows a neuron to selectively attend to inputs coming from its own region. We call the resulting network a segmentation-aware CNN because it adapts its filters at each image point according to local segmentation cues. We demonstrate the merit of our method on two widely different dense prediction tasks, that involve classification (semantic segmentation) and regression (optical flow). Our results show that in semantic segmentation we can match the performance of DenseCRFs while being faster and simpler, and in optical flow we obtain clearly sharper responses than networks that do not use local attention masks. In both cases, segmentation-aware convolution yields systematic improvements over strong baselines. Source code for this work is available online at http://cs.cmu.edu/~aharley/segaware.