Research map / Vision-language and multimodal / Multimodal LLMs
Towards High-Resolution Visual Perception via Hierarchical Entity Exploration
Ziyu Ma, Shidong Yang, Yuxiang Ji, Yiming Hu, Tongwen Huang, Yong Wang, Jianfei Cai, Xiangxiang Chu
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 uniform grid partitioning with entity-based exploration (ablation) | HEE · Visual Probe Hard | accuracy: 37.7 -> 50.0 | Improved |
| add confidence-guided backtracking (ablation) | HEE · Visual Probe Hard | accuracy: 41.5 -> 50.0 | Improved |
| add VLM confidence term M(r,q) to scoring (ablation) | HEE · Visual Probe Hard | accuracy: 21.7 -> 50.0 | Improved |
| add SigLIP similarity term S(q,r) to scoring (ablation) | HEE · Visual Probe Hard | accuracy: 43.4 -> 50.0 | Improved |
| increase entity clusters from 4 to 8 (ablation) | HEE · Visual Probe Hard | accuracy: 50.0 -> 37.7 | Hurt |
| vary lambda from 0.2 to 0.7 vs 0.5 (ablation) | HEE · Visual Probe | average accuracy: 50.7 / 53.2 vs 54.7 | No clear effect |
| use DINO-X instead of YOLO-World detector (ablation) | HEE · Visual Probe Easy | accuracy: 62.4 -> 67.4 (with SigLIP) | Improved |
| use SigLIP instead of CLIP (ablation) | HEE · Visual Probe Easy | accuracy: 62.4 -> 67.4 | Improved |
| apply HEE to Qwen2.5-VL-7B | Qwen2.5-VL-7B · Visual Probe Hard | accuracy: 23.9 -> 50.0 | Improved |
| apply HEE to InternVL2.5-8B | InternVL2.5-8B · HR-Bench 8K Overall | accuracy: 54.8 -> 66.5 | Improved |
| HEE vs RAP and ZoomEye | Qwen2.5-VL-7B · HR-Bench 4K Overall | accuracy and time: 73.9 vs 71.1/71.4; 24min vs 126/46min | Improved |
| apply HEE to Qwen2.5-VL-7B | Qwen2.5-VL-7B · MME-RealWorld Avg | accuracy: 59.99 -> 63.38 | Improved |
Closest papers by meaning
- Divide, Conquer and Combine: A Training-Free Framework for High-Resolution Image Perception in Multimodal Large Language Models (AAAI 2025 · similarity 0.90)
- CVSearch: Empowering Multimodal LLMs with Cognitive Visual Search for High-Resolution Image Perception (ICML 2026 · similarity 0.89)
- BVS: Bayesian Visual Search with Multimodal Large Language Model for Fine-grained Perception (ICML 2026 · similarity 0.89)
- HiDe: Rethinking The Zoom-IN method in High Resolution MLLMs via Hierarchical Decoupling (ICML 2026 · similarity 0.89)
- UVU: Improving Multimodal Understanding via Vision-Language Unified Autoregressive Paradigm (CVPR 2026 · similarity 0.89)
- Seeing What Matters: A Training-Free Self-Guided Framework for Multimodal Detail Perception and Reasoning (CVPR 2026 · similarity 0.88)
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.
