Research map / Vision-language and multimodal / Multimodal LLMs
Global-Local Dual Perception for MLLMs in High-Resolution Text-Rich Image Translation
junxin lu, Tengfei Song, Zhanglin Wu, Lipengfei Lipengfei, Xiaowei Liang, Hui Yang, Kun Chen, NING XIE, Yunfei Lu, Jing Zhao, Shiliang Sun, Daimeng Wei
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 global-local dual perception (GLoTran) to Qwen3-VL 8B | Qwen3-VL 8B · MCiTon En-Zh average | BLEU: 46.7 -> 50.2 | Improved |
| plain SFT of Qwen3-VL 8B on GLoD without the dual-perception design | Qwen3-VL 8B · MCiTon En-Zh average | BLEU: 46.7 -> 46.5 | No clear effect |
| add global-local dual perception to Qwen2.5-VL 7B | Qwen2.5-VL 7B · MCiTon Novel | BLEU: 27.43 -> 44.59 | Improved |
| add global-local dual perception to InternVL2.5 4B | InternVL2.5 4B · MTIT6 ko->zh | BLEU: 2.57 -> 5.97 | Improved |
| keep the global image as context alongside local slices (ablation) | GLoTran · MCiTon News | BLEU: 44.8 -> 57.5 | Improved |
| keep the global image (ablation) | GLoTran · MCiTon Novel | BLEU: 44.1 vs 46.8 | Mixed |
| use text-region detector instead of fixed-size cropping (PSC) (ablation) | GLoTran · MCiTon Novel | BLEU: 36.8 -> 46.8 | Improved |
| add pre-region translation replay (ablation) | GLoTran · MCiTon Novel | BLEU: 45.5 -> 46.8 | Improved |
| feed detector-localized regions to plain MLLMs (ablation) | Qwen3-VL 8B · MCiTon Novel | BLEU: 43.6 -> 42.5 | No clear effect |
| increase global image resolution beyond 448 (ablation) | GLoTran · MCiTon Introduction and Leaflet | BLEU | Hurt |
| set replay window eta to 4 versus smaller or larger (ablation) | GLoTran (Qwen2.5-VL 7B) · MCiTon | BLEU: peak 43.54 at eta=4 | Mixed |
| scale model parameters (InternVL2 8B to 40B) | InternVL2 · MCiTon average | BLEU: 28.1 -> 30.3 | No clear effect |
Closest papers by meaning
- MMTIT-Bench: A Multilingual and Multi-Scenario Benchmark with Cognition-Perception-Reasoning Guided Text-Image Machine Translation (CVPR 2026 · similarity 0.90)
- Divide, Conquer and Combine: A Training-Free Framework for High-Resolution Image Perception in Multimodal Large Language Models (AAAI 2025 · similarity 0.89)
- Kosmos-G: Generating Images in Context with Multimodal Large Language Models (ICLR 2024 · similarity 0.87)
- MARS: Mixture of Auto-Regressive Models for Fine-grained Text-to-image Synthesis (AAAI 2025 · similarity 0.87)
- Zooming without Zooming: Region-to-Image Distillation for Fine-Grained Multimodal Perception (ICML 2026 · similarity 0.87)
- DenseFusion-1M: Merging Vision Experts for Comprehensive Multimodal Perception (NeurIPS 2024 · similarity 0.87)
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.
