Research map / Vision-language and multimodal / Multimodal LLMs
OpenVision: A Fully-Open, Cost-Effective Family of Advanced Vision Encoders for Multimodal Learning
Xianhang Li, Yanqing Liu, Haoqin Tu, Cihang Xie
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 |
|---|---|---|---|
| use patch size 8 instead of 16 (ablation) | OpenVision Ti · LLaVA-1.5 TextVQA | accuracy: 50.2 -> 54.6 | Improved |
| use patch size 8 instead of 16 (ablation) | OpenVision S · LLaVA-1.5 TextVQA | accuracy: 54.3 -> 59.3 | Improved |
| use patch size 8 instead of 16 (ablation) | OpenVision B · LLaVA-1.5 TextVQA | accuracy: 57.9 -> 61.2 | Improved |
| add auxiliary text decoder (ablation) | OpenVision · multimodal benchmarks | accuracy | Improved |
| use synthetic captions instead of original DataComp-1B captions (ablation) | OpenVision · multimodal benchmarks | accuracy | Improved |
| add 336 stage after 224 (resolution progression) (ablation) | OpenVision L/14 · Open-LLaVA-Next TextVQA | accuracy: 64.4 -> 65.7 -> 68.3 | Improved |
| double higher-resolution fine-tuning samples (ablation) | OpenVision L/14 · Open-LLaVA-Next ChartQA | accuracy: 66.1 -> 68.0 | Improved |
| scale Stage 3 data from 665K to 3M (ablation) | OpenVision-Smol · Smol-150M ChartQA | accuracy: 10.8 -> 31.3 | Improved |
| scale input resolution from 384 to 768 (ablation) | OpenVision-Smol · Smol-150M TextVQA | accuracy: 36.3 -> 40.6 | Improved |
| scale Stage 2 data from x1 to x8 (ablation) | OpenVision-Smol · Smol-150M SEED-Bench | accuracy: 50.0 -> 45.1 | No clear effect |
| scale encoder from L/14 to H/14 | OpenVision · LLaVA-1.5 OCR-Bench | score: 315 -> 330 | Improved |
| use OpenVision B/16 instead of OpenAI CLIP B/16 as encoder | Qwen2.5-0.5B LLaVA-1.5 · TextVQA | accuracy: 33.5 -> 34.8 | Improved |
Closest papers by meaning
- Multimodal Autoregressive Pre-training of Large Vision Encoders (CVPR 2025 · similarity 0.89)
- iLLaVA: An Image is Worth Fewer Than 1/3 Input Tokens in Large Multimodal Models (ICLR 2026 · similarity 0.88)
- MoVA: Adapting Mixture of Vision Experts to Multimodal Context (NeurIPS 2024 · similarity 0.87)
- Beyond Encoder Accumulation: Measuring Encoder Roles in Multi-Encoder VLMs (NeurIPS 2026 · similarity 0.86)
- Investigating Redundancy in Multimodal Large Language Models with Multiple Vision Encoders (ICLR 2026 · similarity 0.86)
- Eagle: Exploring The Design Space for Multimodal LLMs with Mixture of Encoders (ICLR 2025 · similarity 0.86)
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.
