README.md

---
annotations_creators:
- expert-generated
language:
- en
license: apache-2.0
size_categories:
- 10K<n<100K
source_datasets:
- original
task_categories:
- image-to-text
- visual-question-answering
- text-generation
pretty_name: Diagram-MMU
tags:
- scientific-diagrams
- diagram-understanding
- diagram-to-code
- tikz
- latex
- code-generation
- diagram-editing
- chart-understanding
- multimodal
- benchmark
configs:
- config_name: diagrams
  data_files:
  - split: test
    path: data/diagrams/test-*.parquet
  - split: testmini
    path: data/diagrams/testmini-*.parquet
- config_name: d2c-p
  data_files:
  - split: test
    path: data/d2c-p/test-*.parquet
  - split: testmini
    path: data/d2c-p/testmini-*.parquet
- config_name: d2c-e
  data_files:
  - split: test
    path: data/d2c-e/test-*.parquet
  - split: testmini
    path: data/d2c-e/testmini-*.parquet
- config_name: dqa
  data_files:
  - split: test
    path: data/dqa/test-*.parquet
  - split: testmini
    path: data/dqa/testmini-*.parquet
dataset_info:
- config_name: diagrams
  features:
  - name: diagram_id
    dtype: string
  - name: image
    dtype: image
  - name: source_code
    dtype: string
  - name: preamble
    dtype: string
  - name: domain
    dtype: string
  splits:
  - name: test
    num_examples: 3744
  - name: testmini
    num_examples: 300
- config_name: d2c-p
  features:
  - name: id
    dtype: string
  - name: diagram_id
    dtype: string
  - name: image
    dtype: image
  - name: instruction
    dtype: string
  - name: code
    dtype: string
  - name: reference_image
    dtype: image
  - name: domain
    dtype: string
  splits:
  - name: test
    num_examples: 3739
  - name: testmini
    num_examples: 300
- config_name: d2c-e
  features:
  - name: id
    dtype: string
  - name: diagram_id
    dtype: string
  - name: image
    dtype: image
  - name: instruction
    dtype: string
  - name: code
    dtype: string
  - name: reference_image
    dtype: image
  - name: domain
    dtype: string
  splits:
  - name: test
    num_examples: 7420
  - name: testmini
    num_examples: 600
- config_name: dqa
  features:
  - name: id
    dtype: string
  - name: diagram_id
    dtype: string
  - name: image
    dtype: image
  - name: question
    dtype: string
  - name: answer
    dtype: string
  - name: domain
    dtype: string
  - name: question_type
    dtype: string
  - name: output_instruction
    dtype: string
  splits:
  - name: test
    num_examples: 7146
  - name: testmini
    num_examples: 600
---

# Diagram-MMU: A Multi-Modal Benchmark for Scientific Diagrams

**ECCV 2026**

🏠 [Project Page](https://vi-ocean.github.io/projects/diagram-mmu) · 💻 [Code](https://github.com/AIGrounding/Diagram-MMU) · 📄 [Paper](https://huggingface.co/papers/2608.12262)

**Diagram-MMU** is a benchmark for evaluating Multimodal Large Language Models (MLLMs) on understanding, parsing, and editing **scientific diagrams**. It contains **3,744** curated diagrams (each with compilable source code) and **18,305** human-validated evaluation instances across **six domains** (`charts`, `planar_geometry`, `3d_shapes`, `graph_structures`, `chemistry`, `circuit_diagrams`), over three tasks: diagram-to-code parsing (**D2C-P**), diagram-to-code editing (**D2C-E**), and diagram question answering (**DQA**).

Each config provides two splits: **`test`** (full benchmark) and **`testmini`** (a class-balanced 50-per-domain subset, 300 diagrams, for quick development). All ground truth is public, so evaluation runs locally with no submission.

| Config     | Task                           | #test | #testmini |
|------------|--------------------------------|------:|----------:|
| `diagrams` | Canonical per-diagram source   | 3,744 | 300 |
| `d2c-p`    | Diagram-to-Code **Parsing**    | 3,739 | 300 |
| `d2c-e`    | Diagram-to-Code **Editing**    | 7,420 | 600 |
| `dqa`      | Diagram **Question Answering** | 7,146 | 600 |

## Data Distribution

<p align="center">
  <img src="assets/data_statistics.png" width="900" alt="Per-domain statistics and task distribution">
</p>

## Main Results

<p align="center">
  <img src="assets/main_results.png" width="900" alt="Main results of 12 MLLMs on Diagram-MMU">
</p>

## Usage

```python
from datasets import load_dataset

d2cp = load_dataset("AIGrounding/Diagram-MMU", "d2c-p", split="test")       # parsing
dqa  = load_dataset("AIGrounding/Diagram-MMU", "dqa", split="testmini")     # QA, dev subset
diagrams = load_dataset("AIGrounding/Diagram-MMU", "diagrams", split="test")

ex = d2cp[0]
ex["image"]   # PIL.Image (decoded automatically)
ex["code"]    # ground-truth source
```

## Evaluation

All ground truth is public, so evaluation runs locally — no submission. The official evaluation code (object / code / image metrics for **D2C-P** & **D2C-E**, and the rule-based + LLM-as-judge pipeline for **DQA**) will be released separately:

- **Evaluation repository:** [github.com/AIGrounding/Diagram-MMU](https://github.com/AIGrounding/Diagram-MMU)

## License & Citation

Released under the **Apache License 2.0**. Source code is collected from official package handbooks (PGFPlots, CircuiTikZ, TKZ-Euclide, ChemFig, TikZ-Network) and community resources (`texample.net`, TeX Stack Exchange, GitHub `tikz_favorites`); upstream sources may carry their own terms. All annotations are original to this work.

```bibtex
@article{bo2026diagrammmu,
  title   = {Diagram-MMU: A Multi-Modal Benchmark for Scientific Diagrams},
  author  = {Bo, Weihao and Zhang, Shan and Sun, Yanpeng and Liu, Jie and Yao, Yongke and Du, Jinhao and He, Wei and Zou, Kai and Li, Zechao and Wang, Jingdong},
  journal = {arXiv preprint arXiv:TODO},
  year    = {2026}
}
```