Learning to Refine Object Segments

TL;DR

This paper introduces SharpMask, a top-down refinement architecture that enhances object segmentation by combining coarse mask encoding with iterative refinement, leading to significant accuracy and speed improvements over previous methods.

Contribution

The paper proposes a novel top-down refinement approach for object segmentation that leverages features at all layers without independent predictions, improving accuracy and efficiency.

Findings

Achieves 10-20% higher average recall than DeepMask.

50% faster processing speed, under 0.8 seconds per image.

Effective refinement of object masks through a simple, fast architecture.

Abstract

Object segmentation requires both object-level information and low-level pixel data. This presents a challenge for feedforward networks: lower layers in convolutional nets capture rich spatial information, while upper layers encode object-level knowledge but are invariant to factors such as pose and appearance. In this work we propose to augment feedforward nets for object segmentation with a novel top-down refinement approach. The resulting bottom-up/top-down architecture is capable of efficiently generating high-fidelity object masks. Similarly to skip connections, our approach leverages features at all layers of the net. Unlike skip connections, our approach does not attempt to output independent predictions at each layer. Instead, we first output a coarse `mask encoding' in a feedforward pass, then refine this mask encoding in a top-down pass utilizing features at successively lower layers. The approach is simple, fast, and effective. Building on the recent DeepMask network for generating object proposals, we show accuracy improvements of 10-20% in average recall for various setups. Additionally, by optimizing the overall network architecture, our approach, which we call SharpMask, is 50% faster than the original DeepMask network (under .8s per image).