Research map / Vision-language and multimodal / Multimodal LLMs
Visual Instruction Tuning
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| visual instruction tuning on GPT-4 generated data (ablation) | LLaVA · LLaVA-Bench (COCO) | relative score: 21.5 -> 85.1 | Helped |
| add detailed description and complex reasoning data to conversation data (ablation) | LLaVA · LLaVA-Bench (COCO) | relative score: 73.8 -> 85.1 | Helped |
| add conversation data to detail + complex (ablation) | LLaVA · LLaVA-Bench (COCO) | relative score: 81.9 -> 85.1 | Helped |
| use CLIP penultimate-layer features instead of last layer (ablation) | LLaVA · ScienceQA | accuracy: 89.96 -> 90.92 | Helped |
| add stage-1 feature alignment pre-training (ablation) | LLaVA · ScienceQA | accuracy: 85.81 -> 90.92 | Helped |
| scale LLM from 7B to 13B (ablation) | LLaVA · ScienceQA | accuracy: 89.84 -> 90.92 | Helped |
| reasoning-first vs answer-first output order (ablation) | LLaVA · ScienceQA | accuracy | No clear effect |
| ensemble with text-only GPT-4 as judge | LLaVA · ScienceQA | accuracy: 90.92 -> 92.53 | Helped |
| GPT-4 complement ensembling | LLaVA · ScienceQA | accuracy: 90.92 -> 90.97 | No clear effect |
| use GPT-4 instead of ChatGPT to generate instruction data (ablation) | LLaVA data pipeline · instruction data quality | qualitative | Helped |
Closest papers by meaning
- Improved Baselines with Visual Instruction Tuning (CVPR 2024 · similarity 0.93)
- VIGC: Visual Instruction Generation and Correction (AAAI 2024 · similarity 0.93)
- LLaDA-V: Large Language Diffusion Models with Visual Instruction Tuning (CVPR 2026 · similarity 0.92)
- Learning to Instruct for Visual Instruction Tuning (NeurIPS 2025 · similarity 0.92)
- VILA: On Pre-training for Visual Language Models (CVPR 2024 · similarity 0.90)
- TaskGalaxy: Scaling Multi-modal Instruction Fine-tuning with Tens of Thousands Vision Task Types (ICLR 2025 · similarity 0.90)
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.