Research map / Vision-language and multimodal / Multimodal LLMs
Griffon v2: Advancing Multimodal Perception with High-Resolution Scaling and Visual-Language Co-Referring
Yufei Zhan, Shurong Zheng, Yousong Zhu, Hongyin Zhao, Fan Yang, Ming Tang, Jinqiao Wang
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 Flamingo-style resampler with strided-conv down-sampling projector (ablation) | Griffon v2 · MSCOCO detection | mAP: 9.6 -> 28.4 | Improved |
| use direct high-resolution encoding instead of image partitioning (Monkey) (ablation) | Griffon v2 · MSCOCO detection (same train data) | mAP: 17.6 -> 24.3 | Improved |
| use direct high-resolution instead of LLaVA-NeXt partition design (ablation) | Griffon v2 · ChartQA | accuracy: 62.2 -> 69.8 | Improved |
| use EVA2-CLIP visual encoder instead of original CLIP-L/14 (ablation) | Griffon v2 · MSCOCO detection | mAP: 29.8 -> 31.9 | Improved |
| use EVA2-CLIP with interpolated positions instead of SAM-CLIP-16 (ablation) | Griffon v2 · MSCOCO detection | mAP: 28.4 -> 31.9 | Improved |
| raise input resolution from 448 to 700 using down-sampling projector (ablation) | Griffon v2 · MSCOCO detection | mAP | Improved |
| increase resolution further with down-sampling projector (ablation) | Griffon v2 · MSCOCO detection | mAP | Improved |
| scale to high resolution with down-sampling projector vs Griffon-13B at 448 | Griffon v2 · MSCOCO val2017 detection | AP_S: 5.9 -> 19.4 | Improved |
| use Griffon v2 instead of Qwen2.5-VL-7B | ODinW · ODinW (13 sets) | avg mAP: 37.0 -> 42.0 | Improved |
| use Griffon v2 instead of ODinW specialist G-DINO-L | ODinW · ODinW (13 sets) | avg mAP: 52.4 -> 42.0 | Hurt |
| use Griffon v2 instead of Lumen (task-specific decoder) | MSCOCO detection · MSCOCO val2017 detection | mAP: 35.3 -> 38.5 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Lumen — Lumen: Unleashing Versatile Vision-Centric Capabilities of Large Multimodal Models (NeurIPS 2024) · 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
- Seeing What Matters: A Training-Free Self-Guided Framework for Multimodal Detail Perception and Reasoning (CVPR 2026 · similarity 0.88)
- Ferret: Refer and Ground Anything Anywhere at Any Granularity (ICLR 2024 · similarity 0.88)
- ClawMachine: Learning to Fetch Visual Tokens for Referential Comprehension (ICLR 2025 · similarity 0.88)
- Divide, Conquer and Combine: A Training-Free Framework for High-Resolution Image Perception in Multimodal Large Language Models (AAAI 2025 · similarity 0.88)
- VisionLLM v2: An End-to-End Generalist Multimodal Large Language Model for Hundreds of Vision-Language Tasks (NeurIPS 2024 · similarity 0.88)
- Perception Tokens Enhance Visual Reasoning in Multimodal Language Models (CVPR 2025 · 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.
