An Empirical Study of Propagation-based Methods for Video Object Segmentation

TL;DR

This paper provides a comprehensive empirical comparison of propagation-based video object segmentation methods, highlighting their performance, design choices, and improvements using a unified framework and ablation studies.

Contribution

It offers the first fair, detailed empirical analysis of propagation-based methods with unified evaluation and improved end-to-end memory networks.

Findings

Achieved a 76.1% mean score on DAVIS 2017 validation set.

Identified key factors influencing segmentation performance.

Demonstrated the effectiveness of improved end-to-end memory networks.

Abstract

While propagation-based approaches have achieved state-of-the-art performance for video object segmentation, the literature lacks a fair comparison of different methods using the same settings. In this paper, we carry out an empirical study for propagation-based methods. We view these approaches from a unified perspective and conduct detailed ablation study for core methods, input cues, multi-object combination and training strategies. With careful designs, our improved end-to-end memory networks achieve a global mean of 76.1 on DAVIS 2017 val set.