Research map / Vision-language and multimodal / Multimodal LLMs
Semantic-Geometric Dual Compression: Training-Free Visual Token Reduction for Ultra-High-Resolution Remote Sensing Understanding
Yueying Li, Fengxiang Wang, Yan Li, Mingshuo Chen, Mengying Zhao, Long Lan
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 geometric stream (IGSR) to semantic stream (ablation) | DualComp · XLRS-Bench | avg accuracy (%): 50.3 -> 53.1 | Improved |
| add semantic stream (SCSA) to geometric stream (ablation) | DualComp · XLRS-Bench | avg accuracy (%): 51.1 -> 53.1 | Improved |
| add explicit topology completion (path tracing) over Top-K selection (ablation) | DualComp · XLRS-Bench | avg accuracy (%): 50.5 -> 53.1 | Improved |
| add topological sequence unrolling over index reordering (ablation) | DualComp · XLRS-Bench | avg accuracy (%): 51.5 -> 53.1 | Improved |
| add text-aware structural modulation (ablation) | DualComp · XLRS-Bench | avg accuracy (%): 51.2 -> 53.1 | Improved |
| add geometric stream IGSR (ablation) | DualComp · XLRS-Bench Route Planning | accuracy (%): 67.0 -> 72.0 | Improved |
| add text-aware modulation (ablation) | DualComp · XLRS-Bench OLUC | accuracy (%): 48.0 -> 53.0 | Improved |
| use DualComp instead of GeoLLaVA-8K static compression | GeoLLaVA-8K · XLRS-Bench | avg accuracy (%): 51.5 -> 53.1 | Improved |
| use DualComp instead of VisionZip at 24x | GeoLLaVA-8K · XLRS-Bench | avg accuracy (%): 42.8 -> 53.1 | Improved |
| use DualComp at 42.4x instead of 24x uncompressed baseline speed | GeoLLaVA-8K · Efficiency | inference speed (s/image): 8.15 -> 3.87 | Improved |
| use DualComp to cut LLM compute | GeoLLaVA-8K · Efficiency | LLM TFLOPs: 198.1 -> 99.8 | Improved |
| apply DualComp to general-purpose Qwen2.5-VL-7B | Qwen2.5-VL-7B · XLRS-Bench (Qwen2.5-VL-7B) | avg accuracy (%): 47.4 -> 47.9 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- GeoLLaVA-8K — GeoLLaVA-8K: Scaling Remote-Sensing Multimodal Large Language Models to 8K Resolution (NeurIPS 2025) · better on 5 of 5 shared results
- +VisionZip — VisionZip: Longer is Better but Not Necessary in Vision Language Models (CVPR 2025) · better on 1 of 1 shared result
- +SparseVLM — SparseVLM: Visual Token Sparsification for Efficient Vision-Language Model Inference (ICML 2025) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- UHR-BAT: Budget-Aware Token Compression Vision-Language model for Ultra-High-Resolution Remote Sensing (ICML 2026 · similarity 0.90)
- GeoLLaVA-8K: Scaling Remote-Sensing Multimodal Large Language Models to 8K Resolution (NeurIPS 2025 · similarity 0.87)
- InstructSAM: A Training-free Framework for Instruction-Oriented Remote Sensing Object Recognition (NeurIPS 2025 · similarity 0.86)
- GeoViS: Geospatially Rewarded Visual Search for Remote Sensing Visual Grounding (CVPR 2026 · similarity 0.86)
- Global Compression Commander: Plug-and-Play Inference Acceleration for High-Resolution Large Vision-Language Models (AAAI 2026 · similarity 0.85)
- ZoomEarth: Active Perception for Ultra-High-Resolution Geospatial Vision-Language Tasks (CVPR 2026 · similarity 0.85)
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.
