Where Do Reported Long-Video QA Gains Come From? A Controlled Audit of Frame Selection, Evidence Injection, and Expert Routing
cs.CV
Submitted: 2026-11-05
Updated: 2026-11-05
Code: https://github.com/mikeslinir/long-video-qa-audit
License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Sources
- HiMu: Hierarchical Multimodal Frame Selection for Long Video Question Answering
- Event-Anchored Frame Selection for Effective Long-Video Understanding
- VideoLLaMA 2: Advancing Spatial-Temporal Modeling and Audio Understanding in Video-LLMs
- Qwen-Audio: Advancing Universal Audio Understanding via Unified Large-Scale Audio-Language Models
- Logic-in-Frames: Dynamic Keyframe Search via Visual Semantic-Logical Verification for Long Video Understanding
- FRAG: Frame Selection Augmented Generation for Long Video and Long Document Understanding
- Adaptive Greedy Frame Selection for Long Video Understanding
- Beyond Frame Selection: Rethinking Long-Video Understanding with MLLMs
- Swift Sampling: Selecting Temporal Surprises via Taylor Series
- When and Where to Look: Adaptive Visual Evidence Scheduling for Efficient Long Video Understanding
- KeyVideoLLM: Towards Large-scale Video Keyframe Selection
- EARS: Evidence-Aware Retrieval via Sound and Motion for Budgeted Video Reasoning
- Beyond Frame Selection: Generative Latent Evidence Aggregation for Long-Video Understanding
- CREST: Curvature-Regulated Event-Centric Sampling for Efficient Long-Video Understanding
- SeViCES: Unifying Semantic-Visual Evidence Consensus for Long Video Understanding
- Qwen2-VL: Enhancing Vision-Language Model's Perception of the World at Any Resolution
- VisualRouter: Query-Grounded Visual Sampling for Long Video Understanding
- Coverage-Driven Adaptive Keyframe Selection for Video Understanding
Related papers
- TEA: Text Encoder Alignment for Robust Concept Erasure in Text-to-Image Models
- Do Egocentric Video-Language Models Capture Both Hand- and Object-Centric Cues?
- Aura: Consistent Multi-Subject Video Generation via VLM-Grounded Semantic Alignment
- PVRF: All-in-one Adverse Weather Removal via Prior-modulated and Velocity-constrained Rectified Flow
- DLWM: Diverse Latent World Models for Efficient Multimodal Reasoning
- Out of Sight, Still in Mind: Token Compression for Omni-LLMs