Research map / Video / Video-language understanding
Decomposed Attention Fusion in MLLMs for Training-free Video Reasoning Segmentation
Su Ho Han, Jeongseok Hyun, Pilhyeon Lee, Minho Shim, Dongyoon Wee, Seon Joo Kim
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 |
|---|---|---|---|
| add contrastive object-background fusion (ablation) | DecAF · Ref-DAVIS attention mask, Qwen2.5VL-7B | J&F: 14.5 -> 25.3 | Improved |
| add contrastive object-background fusion (ablation) | DecAF · ReasonVOS SAM mask, Qwen2.5VL-7B | J&F: 58.4 -> 63.9 | Improved |
| combine video and frame attention (ablation) | DecAF · Ref-DAVIS, Qwen2.5VL-7B | J&F: 65.9 / 67.4 -> 75.2 | Improved |
| add multi-scale frame input (ablation) | DecAF · Ref-DAVIS, InternVL3-8B | J&F: 54.0 -> 62.8 | Improved |
| use V-Max normalized rollout instead of plain rollout (ablation) | DecAF · Ref-DAVIS, Qwen2.5VL-7B | J&F: 68.4 -> 75.2 | Improved |
| add attention consistency score (ablation) | DecAF SAM prompting · Ref-DAVIS, Qwen2.5VL-7B | J&F: 60.0 -> 75.2 | Improved |
| add penalty term to consistency score with Otsu threshold (ablation) | DecAF · Ref-DAVIS, Qwen2.5VL-7B | J&F: 68.3 -> 67.9 | Hurt |
| start rollout from layer 21 instead of 14 (ablation) | DecAF · Ref-DAVIS / ReasonVOS, Qwen2.5VL-7B | J&F: Ref-D 75.2 -> 73.6; ReasV 63.9 -> 64.1 | Mixed |
| raise tau_trk and tau_pq to 0.9 (ablation) | DecAF SAM prompting · ReasonVOS, Qwen2.5VL-7B | J&F: 63.9 -> 66.4 (ReasonVOS); 75.2 -> 74.2 (Ref-DAVIS) | Mixed |
| scale InternVL3 from 8B to 14B | DecAF · ReVOS overall | J&F: 47.4 -> 47.0 | No clear effect |
| scale Qwen2.5VL from 3B to 7B | DecAF · Ref-DAVIS | J&F: 58.1 -> 75.2 | Improved |
| use DecAF instead of Loc-Head with InternVL3-8B | SAM2 prompting · Ref-DAVIS | J&F: 66.3 -> 62.8 | Hurt |
Closest papers by meaning
- The Devil is in Temporal Token: High Quality Video Reasoning Segmentation (CVPR 2025 · similarity 0.88)
- CREMA: Generalizable and Efficient Video-Language Reasoning via Multimodal Modular Fusion (ICLR 2025 · similarity 0.87)
- MVBench: A Comprehensive Multi-modal Video Understanding Benchmark (CVPR 2024 · similarity 0.87)
- Look, Remember and Reason: Grounded Reasoning in Videos with Language Models (ICLR 2024 · similarity 0.87)
- OmniVideoBench: Towards Audio-Visual Understanding Evaluation for Omni MLLMs (ICLR 2026 · similarity 0.86)
- One Token to Seg Them All: Language Instructed Reasoning Segmentation in Videos (NeurIPS 2024 · 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.
