Multi-Agent Target-Existence Verification and Learned Mask Geometry Refinement: Winning Report of the MeViS-Text Track at the 8th LSVOS Challenge 2026

2026-08-11

AI Summary

This work addresses the challenge of spurious detections caused by "object-absent" textual prompts in text-guided video object segmentation. To mitigate this issue, the authors propose SSUPER, a novel framework that first generates candidate mask clips using SAM~3.1 and then employs a multi-agent collaborative mechanism to perform independent existence verification for complete semantic predicates, effectively decoupling existence judgment from segmentation. Furthermore, a StyleRefiner module—trained solely on available training data—is introduced to refine mask geometry to align with annotation conventions without interfering with existence decisions. By integrating heterogeneous multimodal large language model reasoning, multi-agent predicate validation, and mask refinement, SSUPER achieves a score of 0.9081 on the LSVOS 2026 MeViS-Text benchmark, significantly reducing false positives in object-absent scenarios.

Abstract

We present the first-place solution to the MeViS-Text track of the 8th Large-scale Video Object Segmentation (LSVOS) Challenge 2026: referring video object segmentation guided by written motion expressions, including deceptive no-target expressions that match no object in the video and must yield empty masks in every frame. Our pipeline, SSUPER, resolves each expression into a visual concept, generates full-video candidate masklets with SAM~3.1, and selects target IDs. At every reasoning stage, three heterogeneous multimodal large language models independently execute the same stage-specific prompt before a single synthesis pass commits one schema-validated verdict. Although this system rejects every no-target expression in validation, the leaderboard reveals that a substantial share of test no-target cases still slips through. The reason is that hard negatives name a plausible object and fail only under the complete temporal predicate, so when selection and existence are decided together, a category-plausible masklet anchors the verdict. Hence, we decouple existence verification into an independent multi-agent audit of the full predicate (category, count, action, trajectory, event order, and semantic role) that distinguishes absence from temporary invisibility, discounts apparent motion caused by camera movement, and requires contradicting evidence rather than mere uncertainty for a no-target verdict. Without any new segmentation call, this audit recovers most of the residual no-target errors. A training-data-only StyleRefiner then aligns mask geometry with the annotation style of MeViSv2 while preserving every presence decision by construction, showing that once the semantics are fixed, part of the remaining error is stylistic rather than semantic. The complete system reaches a Final score of 0.9081339614 on the official challenge leaderboard.

Problem

• referring video object segmentation
• target-existence verification
• no-target expressions
• mask geometry refinement
• multimodal reasoning

Innovation

• multi-agent verification
• target-existence decoupling
• learned mask refinement
• referring video object segmentation
• multimodal reasoning