Strike the Balance: On-the-Fly Uncertainty based User Interactions for Long-Term Video Object Segmentation

TL;DR

This paper introduces a variant of video object segmentation (VOS) that bridges interactive and semi-automatic approaches, termed Lazy Video Object Segmentation (ziVOS), and aims to maximize the tracking duration of an object of interest, while requiring minimal user corrections to maintain tracking over an extended period.

Abstract

In this paper, we introduce a variant of video object segmentation (VOS) that bridges interactive and semi-automatic approaches, termed Lazy Video Object Segmentation (ziVOS). In contrast, to both tasks, which handle video object segmentation in an off-line manner (i.e., pre-recorded sequences), we propose through ziVOS to target online recorded sequences. Here, we strive to strike a balance between performance and robustness for long-term scenarios by soliciting user feedback's on-the-fly during the segmentation process. Hence, we aim to maximize the tracking duration of an object of interest, while requiring minimal user corrections to maintain tracking over an extended period. We propose a competitive baseline, i.e., Lazy-XMem, as a reference for future works in ziVOS. Our proposed approach uses an uncertainty estimation of the tracking state to determine whether a user interaction is necessary to refine the model's prediction. To quantitatively assess the performance of our method and the user's workload, we introduce complementary metrics alongside those already established in the field. We evaluate our approach using the recently introduced LVOS dataset, which offers numerous long-term videos. Our code is publicly available at https://github.com/Vujas-Eteph/LazyXMem.

Figures (8)

• Figure 1: Visual representation of the aivos framework. (1) The user initiates the segmentation by clicking to identify the object of interest in the video, (2) thus indicating which object to segment. Only when requested by the method (3) does the user provides corrective clicks on-the-fly.
• Figure 2: Overview of Lazy-XMem for Lazy Video Object Segmentation. Our method is relies on an svos baseline (i.e., XMem XMEM). We leverage the entropy to estimate on-the-fly the tracking state. Based on the tracking state's, the method either uses the original mask of the svos baseline, or refine the original mask by generating pseudo-interactions, or requesting user interaction.
• Figure 3: Comparison of correlation coefficients across the DAVIS 2017 Pont-Tuset_arXiv_2017 and LVOS Hong_2023_ICCV datasets: i) the QAM module QDMN and entropy based QDMN QDMN (as Q-$S$ and Q-$S_{\mathcal{R}}$), ii) entropy results for a single baseline (X-$S$, X-$S_{\mathcal{R}}$), an ensemble (E-$S$, E-$S_{\mathcal{R}}$), and Monte-Carlo methods (M-$S$, M-$S_{\mathcal{R}}$), and iii) epistemic uncertainty variants for ensemble and Monte-Carlo (E-$V$, E-$V_{\mathcal{R}}$, M-$V$, M-$V_{\mathcal{R}}$).
• Figure S1: Qualitative results on the validation set of LVOS Hong_2023_ICCV when refining the mask through pseudo-corrections (Success cases).
• Figure S2: Qualitative results on the validation set of LVOS Hong_2023_ICCV when refining the mask through pseudo-corrections (Failure cases).