MiNQVIS: Mitigating Noisy Queries for Robust Online Video Instance Segmentation

1School of Computer Science, Wuhan University, 2Leibniz Institute for Analytical Sciences (ISAS)
*Corresponding author
Overview of MiNQVIS with query-level gating, video-level prototypical contrastive learning, and dual-memory association.

MiNQVIS mitigates noisy instance queries throughout training and inference to stabilize representation learning and online association.

Abstract

Online Video Instance Segmentation (VIS) relies on per-frame instance queries for causal detection, segmentation, and temporal association. However, in realistic videos with occlusion, blur, and clutter, these predictions often become noisy, exhibiting false positives, missed instances, or inaccurate masks. Such noisy queries destabilize cross-frame association, contaminate contrastive supervision, and propagate drift through memory-based matching.

We introduce MiNQVIS, the first noise-aware framework for online VIS that stabilizes both training and inference. It combines query-level gating with embedding-norm stabilization, video-level prototypical contrastive learning (VPCL) from reliable matches, and dual-memory association that fuses momentum embeddings with history-best retrieval. Experiments on YouTubeVIS 2019/2021 and OVIS show consistent improvements over strong online VIS baselines, with large gains in challenging scenes.

Method

MiNQVIS builds on a standard Mask2Former-based online VIS pipeline and introduces three complementary components to prevent unreliable queries from affecting supervision and temporal association.

Query-level gating

Classification-guided thresholding removes low-confidence queries in both training and inference. L2-normalized embeddings decouple similarity from feature magnitude.

Video-level prototypes

VPCL aggregates reliable cross-frame matches into stable instance prototypes, reducing the influence of noisy frames on contrastive supervision.

Dual-memory association

Momentum memory captures short-term continuity, while history-best retrieval preserves a reliable reference when recent observations are corrupted.

Comparison of previous norm-based suppression and MiNQVIS confidence-guided query gating.

Query-level gating explicitly filters unreliable predictions and stabilizes similarity through embedding normalization.

Qualitative Results

Qualitative comparison of CTVIS, CAVIS, and MiNQVIS under occlusion-induced fragmentation.

Under heavy occlusion, MiNQVIS maintains cleaner masks and more consistent identities across frames.

Main Results

MiNQVIS is evaluated on YouTubeVIS 2019, YouTubeVIS 2021, and OVIS with ResNet-50 and Swin-L backbones. It achieves the best OVIS AP under both backbones, demonstrating robust association in crowded and heavily occluded scenes.

Backbone YTVIS19 AP YTVIS21 AP OVIS AP
ResNet-50 55.6 51.7 38.2
Swin-L 66.2 61.6 49.1

Component Ablation

Query-level gating VPCL Dual-memory fusion YTVIS21 AP OVIS AP
47.534.7
49.835.9
48.736.3
48.135.5
51.437.8
51.738.2

Gating Analysis

Most filtered queries are false positives or fragmented masks. In this OVIS example, the gate removes 193 unreliable queries while retaining all valid foreground instances.

Qualitative analysis and composition of queries filtered by MiNQVIS.

Failure Cases

MiNQVIS failure cases under extreme motion and pose ambiguity.

Extreme motion and drastic pose changes can still cause temporary identity confusion or missed instances.

BibTeX

@article{qiao2026minqvis,
  author = {Qiao, Jie and Chen, Jianxu and Xu, Xiaowei},
  title  = {MiNQVIS: Mitigating Noisy Queries for Robust Online Video Instance Segmentation},
  year   = {2026}
}