Research map / Vision-language and multimodal / Multimodal LLMs
Oasis: One Image is All You Need for Multimodal Instruction Data Synthesis
Letian Zhang, Quan Cui, Bingchen Zhao, Cheng Yang
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 Oasis synthetic data to LLaVA-NeXT SFT data | LLaVA-NeXT Vicuna-7B-v1.5 · 14-benchmark average | average score: 53.0 -> 56.1 | Improved |
| add Oasis synthetic data to LLaVA-NeXT SFT data | LLaVA-NeXT Qwen2.5-7B · 14-benchmark average | average score: 61.4 -> 63.2 | Improved |
| add Oasis synthetic data to LLaVA-NeXT SFT data | LLaVA-NeXT Llama3-8B · 14-benchmark average | average score: 55.8 -> 59.0 | Improved |
| add Oasis data in place of Cambrian original annotations | Vicuna-7B-v1.5 · 14-benchmark average | average score: 54.9 -> 56.1 | Improved |
| add Oasis data instead of MMEvol data of equal size | Vicuna-7B-v1.5 · 14-benchmark average | average score: 51.6 -> 56.1 | Improved |
| scale Oasis data from 0 to 150k added to 100k SFT | LLaVA-100k · 14-benchmark average | average score: 46.5 -> 46.6 | No clear effect |
| scale Oasis data from 0 to 500k added to 100k SFT | LLaVA-100k · 14-benchmark average | average score: 46.5 -> 51.7 | Improved |
| add Oasis data to baseline | Vicuna-7B-v1.5 · DocVQA | accuracy: 71.7 -> 76.0 | Improved |
| add 70k OCR-domain Oasis data | LLaVA-NeXT Vicuna-7B-v1.5 · DocVQA | accuracy: 71.7 -> 75.0 | Improved |
| add 2k medical Oasis data to 2k LLaVA data | LLaVA-NeXT · PathVQA | accuracy: 34.0 -> 41.2 | Improved |
| add 250k recycled caption data (ablation) | Oasis · 14-benchmark average | average score: 56.1 -> 56.4 | Improved |
| add instruction quality control (ablation) | Oasis (200k) · 14-benchmark average | average score: 53.9 -> 54.9 | Improved |
Closest papers by meaning
- MIA-Bench: Towards Better Instruction Following Evaluation of Multimodal LLMs (ICLR 2025 · similarity 0.88)
- NaViL: Rethinking Scaling Properties of Native Multimodal Large Language Models under Data Constraints (NeurIPS 2025 · similarity 0.86)
- VIGC: Visual Instruction Generation and Correction (AAAI 2024 · similarity 0.86)
- Web2Code: A Large-scale Webpage-to-Code Dataset and Evaluation Framework for Multimodal LLMs (NeurIPS 2024 · similarity 0.86)
- MM1.5: Methods, Analysis & Insights from Multimodal LLM Fine-tuning (ICLR 2025 · similarity 0.86)
- Pink: Unveiling the Power of Referential Comprehension for Multi-modal LLMs (CVPR 2024 · 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.
