Getting to 99% Accuracy in Interactive Segmentation

TL;DR

This paper introduces a new interactive segmentation architecture and training scheme that significantly improves accuracy, reaching 99%, by better leveraging user interactions and synthetic datasets designed for complex boundaries.

Contribution

The paper presents a novel interactive segmentation architecture and training scheme tailored to exploit user interactions more effectively and introduces a synthetic dataset for complex object boundaries.

Findings

Achieved state-of-the-art performance in interactive segmentation.

Significant accuracy improvements up to 99%.

Enhanced utilization of user interactions and synthetic data.

Abstract

Interactive object cutout tools are the cornerstone of the image editing workflow. Recent deep-learning based interactive segmentation algorithms have made significant progress in handling complex images and rough binary selections can typically be obtained with just a few clicks. Yet, deep learning techniques tend to plateau once this rough selection has been reached. In this work, we interpret this plateau as the inability of current algorithms to sufficiently leverage each user interaction and also as the limitations of current training/testing datasets. We propose a novel interactive architecture and a novel training scheme that are both tailored to better exploit the user workflow. We also show that significant improvements can be further gained by introducing a synthetic training dataset that is specifically designed for complex object boundaries. Comprehensive experiments support our approach, and our network achieves state of the art performance.