Research map / Vision-language and multimodal / Multimodal LLMs
Visual instruction tuning
Fine-tunes language models on image-instruction data, generated or filtered, to follow multimodal instructions.
68 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2023–2026) · an approach in Multimodal LLMs, keyword cluster “mllm · instruction · llms”
State History Benchmarks Tried Checklist Preprints Who Export
Where it stands
Leveling off Several papers from 2025–26 still report on 1 tracked benchmark, but the best result has barely moved since 2024.
24 papers in 2023–24 and 44 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic shrank, from 11.6% to 2.0%.
What looks settled
- POPE (score): best up to 2024 88.7, best in 2025–26 88.1 from 2 papers, at most 0.2% better, if at all.
How this is decided
A benchmark is tracked when at least four papers of this approach report their own method on the same dataset and metric in at least two different years. Its record is the best number reported up to 2024 compared with the best reported in 2025–26: still improving means at least 1% better, flat means at most 0.2% better with at least two papers from 2025–26 trying. Numbers on a different scale from the rest (a fraction among percentages) and far outliers are left out. Results come from different papers with different setups, so they are signals, not head-to-head comparisons. The share of papers is counted at ICML, NeurIPS, ICLR, CVPR and AAAI.
How it got here
Most cited paper of each year
- 2023 · Visual Instruction Tuning (NeurIPS 2023 · 1,198 citations)
- 2024 · Improved Baselines with Visual Instruction Tuning (CVPR 2024 · 1,490 citations)
- 2025 · FireEdit: Fine-grained Instruction-based Image Editing via Region-aware Vision Language Model (CVPR 2025 · 10 citations)
- 2026 · LLaDA-V: Large Language Diffusion Models with Visual Instruction Tuning (CVPR 2026 · 1 citation)
Earliest papers here
- Visual Instruction Tuning (NeurIPS 2023 · 1,198 citations)
- Cheap and Quick: Efficient Vision-Language Instruction Tuning for Large Language Models (NeurIPS 2023 · 45 citations)
- LAMM: Language-Assisted Multi-Modal Instruction-Tuning Dataset, Framework, and Benchmark (NeurIPS 2023 · 40 citations)
Benchmark records
The best number each year that beat every earlier one, among papers of this approach reporting their own method. Numbers are extracted by AI from each paper's text; check the paper before citing.
POPE · score (higher is better)
4 papers, 2024–2025 · flat since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2024 | 88.7 | Meteor: Mamba-based Traversal of Rationale for Large Language and Vision Models (NeurIPS 2024 · 0 citations) |
VQAv2 · accuracy (higher is better)
4 papers, 2024–2025 · one report since 2024, not better
| Year | Best so far | Set by |
|---|---|---|
| 2024 | 85.1 | PerceptionGPT: Effectively Fusing Visual Perception into LLM (CVPR 2024 · 36 citations) |
What has been tried
48 papers here report 575 measured changes (a component added, removed or swapped, compared with a baseline): helped 73%, no clear effect 4%, hurt 18%, mixed 5%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| replace linear vision-language projection with a two-layer MLP connector | Academic and instruction-following benchmarks (GQA, MME, MM-Vet) · score | MME 1197.0 -> 1355.2 (vs +VQA-v2 row); MM-Vet 27.7 -> 27.8 region-level context | Helped | Improved Baselines with Visual Instruction Tuning (CVPR 2024 · 1,490 citations) |
| Add BIM instead of using Plain | Reason-Edit understanding scenarios · Ins-align | 0.695 -> 0.712 | Helped | SmartEdit: Exploring Complex Instruction-based Image Editing with Multimodal Large Language Models (CVPR 2024 · 71 citations) |
| Increase Stage 1 training from 1 to 3 epochs | Quilt-VQA · Open recall | 59.82 -> 60.97 | Helped | Quilt-LLaVA: Visual Instruction Tuning by Extracting Localized Narratives from Open-Source Histopathology Videos (CVPR 2024 · 54 citations) |
| use continuous token embeddings instead of numerical box tokens with LoRA rank 8 | RefCOCO val (limited training data) · Acc@0.5 | 0.79 -> 69.7 | Helped | PerceptionGPT: Effectively Fusing Visual Perception into LLM (CVPR 2024 · 36 citations) |
| add Categorical instruction data | EmoSet test · accuracy (%) | 42.20 -> 80.90 | Helped | EmoVIT: Revolutionizing Emotion Insights with Visual Instruction Tuning (CVPR 2024 · 30 citations) |
| add VIGC-generated data corrected by VIC (vs. uncorrected VIG) for detail description | LLaVA-Eval · All relative score | 81.4 -> 85.8 | Helped | VIGC: Visual Instruction Generation and Correction (AAAI 2024 · 28 citations) |
| Use clinician preference alignment in both stages rather than neither | Open-ended medical visual chat (overall) · GPT-4V score | 64.15 -> 68.63 | Helped | Biomedical Visual Instruction Tuning with Clinician Preference Alignment (NeurIPS 2024 · 20 citations) |
| adaptive contour sampling (gradient-aware) instead of uniform sampling for mask sequences | RefCOCO val, upper bound with 32 points · mIoU upper bound | 97.26 vs 94.70 | Helped | Jack of All Tasks Master of Many: Designing General-Purpose Coarse-to-Fine Vision-Language Model (CVPR 2024 · 17 citations) |
| Polite Flamingo rewriting of raw annotations into PF-1M used for visual instruction tuning (vs raw annotations) | Ablation Figure (held-in/held-out accuracy and reward) · politeness (reward score) | Helped | Visual Instruction Tuning with Polite Flamingo (AAAI 2024 · 13 citations) | |
| Add masking to the elected unified task vector | 8 vision tasks, ViT-B/32 · Avg Acc | 59.3 -> 86.8; [\uparrow 27.5] | Helped | EMR-Merging: Tuning-Free High-Performance Model Merging (NeurIPS 2024 · 13 citations) |
| Include instruction loss to train concise expressive instructions | MA5k · SSIM | 58.18 -> 66.25 | Helped | Guiding Instruction-based Image Editing via Multimodal Large Language Models (ICLR 2024 · 11 citations) |
| add region-aware mixed-modal encoding (region tokens) | Emu Edit test · L1 | 0.0897 -> 0.0859 | Helped | FireEdit: Fine-grained Instruction-based Image Editing via Region-aware Vision Language Model (CVPR 2025 · 10 citations) |
What reviewers will expect
Datasets and baselines named in the 44 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- COCO · 5 of 44 papers
- VQA-v2 · 4 of 44 papers
- LLaVA-665K · 4 of 44 papers
- LCS-558K · 4 of 44 papers
- OKVQA · 3 of 44 papers
- Vision-Flan · 3 of 44 papers
- MagicBrush · 3 of 44 papers
- ReasonEdit · 3 of 44 papers
Baselines
- GPT-4o · 6 of 44 papers
- InstructPix2Pix · 5 of 44 papers
- MagicBrush · 5 of 44 papers
- MGIE · 4 of 44 papers
- BLIP-2 · 4 of 44 papers
- MEND · 3 of 44 papers
- Random · 3 of 44 papers
- CLIP-Score · 3 of 44 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 68 papers; from OpenAlex and the papers' affiliations).
- The Chinese University of Hong Kong · 6 papers
- Hong Kong University of Science and Technology · 5 papers
- SenseTime · 4 papers
- Shanghai AI Laboratory · 4 papers
- Tsinghua University · 4 papers
- Zhejiang University · 4 papers
- Fudan University · 4 papers
- Alibaba Group · 3 papers
- University of Wisconsin - Madison · 3 papers
- University of Washington · 3 papers
Use it
Download BibTeX of 23 key papersOpen the topic in the interactive map
Get alerts. Describe your idea in scime atlas and save it (free scime account): you get a weekly email when new preprints land close to it.
Related approaches
Same keyword cluster
- Instruction-tuned multimodal assistants (199 papers)
- Benchmarks and analysis of MLLMs (181 papers)
- Unified any-modality language models (152 papers)
- Interleaved image-text generation with MLLMs (153 papers)
Elsewhere in the topic
- Chain-of-thought multimodal reasoning (332 papers)
- VLM capability analysis (264 papers)
- Multimodal LLM benchmarks (252 papers)
- Spatial reasoning in VLMs (207 papers)
- Reinforcement learning for visual reasoning (163 papers)
- Attention-based visual token pruning (117 papers)
- Decoding-time hallucination mitigation (91 papers)
- Training-free token reduction for MLLMs (90 papers)
Benchmark numbers, ablations and baseline numbers are extracted by AI from each paper's full text and can be wrong; the paper is the authority. Data: results index of 2026-11-02, citation counts from OpenAlex (2026-10-09).