Temporally stable video segmentation without video annotations

TL;DR

This paper presents an unsupervised method to adapt still image segmentation models for stable, temporally consistent video segmentation by leveraging optical flow and a consistency measure validated against human judgment.

Contribution

It introduces a novel unsupervised approach combining optical flow-based consistency with a multi-input decoder to improve video segmentation stability without video annotations.

Findings

Enhanced temporal stability in video segmentation results

Minimal loss of accuracy compared to image-based models

Validated consistency measure correlates well with human judgment

Abstract

Temporally consistent dense video annotations are scarce and hard to collect. In contrast, image segmentation datasets (and pre-trained models) are ubiquitous, and easier to label for any novel task. In this paper, we introduce a method to adapt still image segmentation models to video in an unsupervised manner, by using an optical flow-based consistency measure. To ensure that the inferred segmented videos appear more stable in practice, we verify that the consistency measure is well correlated with human judgement via a user study. Training a new multi-input multi-output decoder using this measure as a loss, together with a technique for refining current image segmentation datasets and a temporal weighted-guided filter, we observe stability improvements in the generated segmented videos with minimal loss of accuracy.