Temporally stable video segmentation without video annotations
This addresses the challenge of scarce video annotations for researchers and practitioners in video analysis, offering an incremental improvement by leveraging existing image datasets.
The paper tackles the problem of adapting still image segmentation models to video without requiring video annotations, using an optical flow-based consistency measure and a new decoder to improve temporal stability. The method achieves stability improvements in segmented videos with minimal loss of accuracy, as verified by a user study.
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.