GIFT: Global Irreplaceability Frame Targeting for Efficient Video Understanding

arXiv:2603.25072 · cs.CV · Submitted 2026-03-26 · Read on arXiv

cs.CV

Submitted: 2026-03-26

Updated: 2026-10-11

Comments: 11 pages, 3 figures. Accepted by CVPR 2026. Code: https://github.com/ylnxxts/GIFT

Code: https://github.com/ylnxxts/GIFT

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

Terminology

• Benchmark — 1× in this paper · explained in 2 episodes
• Core — 1× in this paper · explained in 2 episodes
• Training-Free Framework — 1× in this paper · explained in 2 episodes

Sources

• Qwen2.5-VL Technical Report
• Variation-aware Vision Token Dropping for Faster Large Vision-Language Models
• LongVILA: Scaling Long-Context Visual Language Models for Long Videos
• Expanding Performance Boundaries of Open-Source Multimodal Models with Model, Data, and Test-Time Scaling
• Scaling Video-Language Models to 10K Frames via Hierarchical Differential Distillation
• OmniSIFT: Modality-Asymmetric Token Compression for Efficient Omni-modal Large Language Models
• Video-MME: The First-Ever Comprehensive Evaluation Benchmark of Multi-modal LLMs in Video Analysis
• Logic-in-Frames: Dynamic Keyframe Search via Visual Semantic-Logical Verification for Long Video Understanding
• FrameThinker: Learning to Think with Long Videos via Multi-Turn Frame Spotlighting
• LLaVA-OneVision: Easy Visual Task Transfer
• VideoChat: Chat-Centric Video Understanding
• MaxInfo: A Training-Free Key-Frame Selection Method Using Maximum Volume for Enhanced Video Understanding
• Video-LLaVA: Learning United Visual Representation by Alignment Before Projection
• Mixing Importance with Diversity: Joint Optimization for KV Cache Compression in Large Vision-Language Models
• MMG-Vid: Maximizing Marginal Gains at Segment-level and Token-level for Efficient Video LLMs
• LongVU: Spatiotemporal Adaptive Compression for Long Video-Language Understanding
• From Frames to Clips: Training-free Adaptive Key Clip Selection for Long-Form Video Understanding
• MDP3: A Training-free Approach for List-wise Frame Selection in Video-LLMs
• Qwen2-VL: Enhancing Vision-Language Model's Perception of the World at Any Resolution
• Accelerating Streaming Video Large Language Models via Hierarchical Token Compression

Related papers

• Loss Knows Best: Detecting Annotation Errors in Videos via Loss Trajectories
• AnchorWeave: World-Consistent Video Generation with Retrieved Local Spatial Memories
• Benchmarking the Robustness of Foundation Models for Mammography under Domain Shift
• MambaX-Net: Dual-Input Mamba-Enhanced Cross-Attention Network for Longitudinal MRI Segmentation
• TeleOCR: Navigating Document Parsing Across Digital and Camera-Captured Documents
• A Survey on Efficient Vision-Language-Action Models