DIVE: Dynamic Iterative Visual Evidence Construction for Efficient Vision-Language Models
Chen Zhong, Xiao An, Zijie Wang, Jiepan Li, Guangyi Yang, Wei He
cs.CV, cs.LG
Submitted: 2026-08-05
Code: https://github.com/Zhong-Chenchen/DIVE
License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Terminology
Sources
- Token Merging: Your ViT But Faster
- SCOPE: Saliency-Coverage Oriented Token Pruning for Efficient Multimodel LLMs
- MMTok: Multimodal Coverage Maximization for Efficient Inference of VLMs
- Prune Redundancy, Preserve Essence: Vision Token Compression in VLMs via Synergistic Importance-Diversity
- Video-MME: The First-Ever Comprehensive Evaluation Benchmark of Multi-modal LLMs in Video Analysis
- Seed1.5-VL Technical Report
- LLaVA-OneVision: Easy Visual Task Transfer
- LLaVA-NeXT-Interleave: Tackling Multi-image, Video, and 3D in Large Multimodal Models
- MVBench: A Comprehensive Multi-modal Video Understanding Benchmark
- ResPrune: Text-Conditioned Subspace Reconstruction for Visual Token Pruning in Large Vision-Language Models
- Video-XL-Pro: Reconstructive Token Compression for Extremely Long Video Understanding
- OCRBench: On the Hidden Mystery of OCR in Large Multimodal Models
- ViTCoP: Accelerating Large Vision-Language Models via Visual and Textual Semantic Collaborative Pruning
- Qwen2-VL: Enhancing Vision-Language Model's Perception of the World at Any Resolution
- Rethinking Token Reduction for Large Vision-Language Models
- PyramidDrop: Accelerating Your Large Vision-Language Models via Pyramid Visual Redundancy Reduction
- VideoLLaMA 3: Frontier Multimodal Foundation Models for Image and Video Understanding
- Efficient Inference for Large Vision-Language Models: Bottlenecks, Techniques, and Prospects
- Beyond Attention or Similarity: Maximizing Conditional Diversity for Token Pruning in MLLMs
- LLaVA-Video: Video Instruction Tuning With Synthetic Data
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