Research map / Video / Video segmentation and tracking
Training-Free Spatio-temporal Decoupled Reasoning Video Segmentation with Adaptive Object Memory
Zhengtong Zhu, Jiaqing Fan, Zhixuan Liu, Fanzhang Li
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace first-frame sampling with global sampling (ablation) | SDAM · Ref-DAVIS17 val | J&F: 69.2 -> 73.4 | Improved |
| add Motion Driven Sampler over global sampling (ablation) | SDAM · Ref-DAVIS17 val | J&F: 73.4 -> 76.0 | Improved |
| add Motion Driven Sampler over global sampling (ablation) | SDAM · MeViS val | J&F: 54.4 -> 55.8 | Improved |
| raise MLLM confidence weight a from 0.4 to 0.75 (ablation) | SDAM · Ref-DAVIS17 val | J&F: 73.9 -> 76.0 | Improved |
| raise MLLM confidence weight a from 0.4 to 0.6, then 0.75 (ablation) | SDAM · ReasonVOS | J&F: 54.1 -> 55.1 (a=0.6) -> 54.7 (a=0.75) | Mixed |
| use top-n keyframes (n=2,3) instead of a single keyframe (ablation) | SDAM · Ref-DAVIS17 val (global sampling) | J&F: 73.4 -> 70.7 (n=2), 68.6 (n=3) | Hurt |
| training-free SDAM versus trained DMVS/SSA | SDAM · MeViS val | J&F: 48.6 vs 48.6 | No clear effect |
| training-free SDAM versus best prior method | SDAM · Ref-DAVIS17 val | J&F: 70.4 -> 76.0 | Improved |
| training-free SDAM versus GLUS | SDAM · ReasonVOS | J&F: 49.9 -> 55.1 | Improved |
| training-free SDAM versus GLUS | SDAM · ReVOS val (Overall) | J&F: 54.9 -> 58.0 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- GLUS — GLUS: Global-Local Reasoning Unified into A Single Large Language Model for Video Segmentation (CVPR 2025) · better on 4 of 4 shared results
- VISA — VISA: Reasoning Video Object Segmentation via Large Language Model (ECCV 2024) · better on 4 of 4 shared results
- InstructSeg — InstructSeg: Unifying Instructed Visual Segmentation with Multi-modal Large Language Models (ICCV 2025) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- The Devil is in Temporal Token: High Quality Video Reasoning Segmentation (CVPR 2025 · similarity 0.89)
- One Token to Seg Them All: Language Instructed Reasoning Segmentation in Videos (NeurIPS 2024 · similarity 0.88)
- Fast Reasoning Segmentation for Images and Videos (CVPR 2026 · similarity 0.87)
- Unleashing the Temporal-Spatial Reasoning Capacity of GPT for Training-Free Audio and Language Referenced Video Object Segmentation (AAAI 2025 · similarity 0.87)
- DTOS: Dynamic Time Object Sensing with Large Multimodal Model (CVPR 2025 · similarity 0.86)
- Learning Spatial-Semantic Features for Robust Video Object Segmentation (ICLR 2025 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
