MathGen / README.md
arielyao's picture
update
7180561
|
Raw
History Blame Contribute Delete
5.77 kB
metadata
pretty_name: MathGen
license: mit
language:
  - en
task_categories:
  - text-to-image
tags:
  - mathematics
  - text-to-image
  - image-generation
  - benchmark
  - script-as-a-judge
  - mathematical-reasoning
size_categories:
  - n<1K
configs:
  - config_name: default
    data_files:
      - split: clean_scene
        path: data/clean_scene.jsonl
      - split: open_scene
        path: data/open_scene.jsonl

MathGen

Revealing the Illusion of Mathematical Competence through Text-to-Image Generation

arXiv Code Project Page

Introduction

MathGen is a benchmark for evaluating mathematical correctness in text-to-image generation. It contains 420 problems across seven core mathematical domains: counting, angle, fraction, function, plane geometry, set, and solid geometry.

The benchmark includes a 350-problem Clean-Scene set and a 70-problem Open-Scene set with paired mathematical constraints. Each problem is evaluated with deterministic, prompt-conditioned verification scripts rather than subjective visual preference.

MathGen seven mathematical domains

Dataset Structure

This repository provides two splits:

Split Rows Description
clean_scene 350 Controlled mathematical diagram prompts across seven domains.
open_scene 70 Realistic-scene prompts paired with Clean-Scene mathematical constraints.

Each row contains:

Column Description
id Stable problem id.
split clean_scene or open_scene.
topic Mathematical domain or open-scene category.
subtopic Fine-grained task type.
prompt Text-to-image prompt.
original_id Original local problem id.
source_topic Source mathematical domain.

Evaluation Pipeline

MathGen benchmark and evaluation pipeline

MathGen uses a Script-as-a-Judge protocol. Generated images are checked by deterministic verification scripts for numerical, geometric, topological, symbolic, and functional constraints.

Leaderboard

Main results on the 350-problem Clean-Scene set. Accuracy is reported across seven mathematical domains, with 50 problems per domain. Best results are shown in bold, and second-best results are shown with underline.

Model Counting Angle Fraction Function Plane Set Solid Overall
SD-3-Medium 0.0 0.0 0.0 0.0 6.0 0.0 6.0 1.7
SD-3.5-Medium 4.0 0.0 0.0 0.0 12.0 2.0 6.0 3.4
SD-3.5-Large 12.0 0.0 2.0 0.0 12.0 2.0 4.0 4.6
FLUX-2 8.0 2.0 8.0 2.0 42.0 8.0 8.0 11.1
PixArt-Sigma 10.0 0.0 2.0 0.0 12.0 2.0 8.0 4.9
PixArt-XL-2 0.0 0.0 0.0 0.0 14.0 0.0 4.0 2.6
HiDream-I1 6.0 0.0 0.0 2.0 4.0 2.0 6.0 2.9
Qwen-Image 22.0 0.0 8.0 2.0 24.0 4.0 6.0 9.4
Z-Image-Turbo 8.0 0.0 8.0 0.0 16.0 2.0 14.0 6.9
Infinity-8B 6.0 0.0 4.0 0.0 18.0 2.0 8.0 5.4
GoT-R1-7B 8.0 0.0 0.0 0.0 16.0 2.0 2.0 4.0
BAGEL 4.0 0.0 0.0 0.0 14.0 0.0 2.0 2.9
show-o2-1.5B 0.0 0.0 0.0 4.0 18.0 0.0 2.0 3.4
show-o2-7B 0.0 0.0 0.0 4.0 10.0 0.0 8.0 3.1
Janus-Pro-1B 0.0 0.0 0.0 0.0 14.0 0.0 2.0 2.3
Janus-Pro-7B 0.0 0.0 0.0 0.0 12.0 2.0 6.0 2.9
BLIP3o-4B 4.0 0.0 2.0 0.0 14.0 2.0 8.0 4.3
BLIP3o-8B 6.0 0.0 2.0 0.0 14.0 2.0 4.0 4.0
OmniGen2-7B 4.0 0.0 2.0 0.0 12.0 2.0 8.0 4.0
FLUX-2-Pro 22.0 10.0 20.0 18.0 54.0 16.0 20.0 22.9
FLUX-Kontext-Pro 10.0 0.0 10.0 4.0 18.0 6.0 4.0 7.4
Seedream 3.0 14.0 0.0 2.0 0.0 24.0 6.0 8.0 7.7
Seedream 4.0 20.0 0.0 6.0 10.0 36.0 6.0 14.0 13.1
Ideogram v3 Turbo 10.0 2.0 0.0 0.0 20.0 2.0 8.0 6.0
Nano Banana 20.0 8.0 24.0 10.0 64.0 10.0 24.0 22.9
Nano Banana Pro 48.0 54.0 50.0 70.0 72.0 42.0 40.0 53.7
Imagen 4 12.0 2.0 2.0 2.0 12.0 0.0 12.0 6.0
Imagen 4 Ultra 20.0 6.0 16.0 4.0 42.0 8.0 16.0 16.0
GPT-Image-1 32.0 12.0 44.0 16.0 68.0 20.0 24.0 30.9
GPT-Image-1.5 56.0 24.0 54.0 22.0 70.0 20.0 28.0 39.1

Examples

MathGen qualitative examples

Citation

@misc{liu2026mathgen,
  title={MathGen: Revealing the Illusion of Mathematical Competence through Text-to-Image Generation},
  author={Liu, Ruiyao and Shen, Hui and Zhang, Ping and Hsieh, Yunta and Zhang, Yifan and Xu, Jing and Han, Qi and Li, Junchen and Lu, Jiawei and Ma, Jianing and Mo, Jiaqi and Chen, Sicheng and Zhang, Zhen and Wan, Zhongwei and Xiong, Jing and Wang, Xin and Liu, Ziyuan and Cao, Hangrui and Wong, Ngai},
  year={2026},
  eprint={2603.27959},
  archivePrefix={arXiv},
  primaryClass={cs.CV},
  url={https://arxiv.org/abs/2603.27959}
}