Datasets:

arielyao
/

MathGen

id stringlengths 7 12	split stringclasses 1 value	topic stringclasses 7 values	subtopic stringclasses 19 values	prompt stringlengths 25 649	original_id int64 1 99	source_topic stringclasses 8 values
angle_001	clean_scene	angle	angle_measurement	Draw one acute angle of 25°.	1	angle
angle_002	clean_scene	angle	angle_measurement	Draw one acute angle of 35°.	2	angle
angle_003	clean_scene	angle	angle_measurement	Draw one acute angle of 40°.	3	angle
angle_004	clean_scene	angle	angle_measurement	Draw one acute angle of 45°.	4	angle
angle_005	clean_scene	angle	angle_measurement	Draw one acute angle of 50°.	5	angle
angle_006	clean_scene	angle	angle_measurement	Draw one acute angle of 60°.	6	angle
angle_007	clean_scene	angle	angle_measurement	Draw one acute angle of 65°.	7	angle
angle_008	clean_scene	angle	angle_measurement	Draw one acute angle of 70°.	8	angle
angle_009	clean_scene	angle	angle_measurement	Draw one acute angle of 75°.	9	angle
angle_010	clean_scene	angle	angle_measurement	Draw one acute angle of 80°.	10	angle
angle_011	clean_scene	angle	angle_measurement	Draw one right angle (90°). Mark it with the standard square symbol.	11	angle
angle_012	clean_scene	angle	angle_measurement	Draw one obtuse angle of 100°.	12	angle
angle_013	clean_scene	angle	angle_measurement	Draw one obtuse angle of 110°.	13	angle
angle_014	clean_scene	angle	angle_measurement	Draw one obtuse angle of 120°.	14	angle
angle_015	clean_scene	angle	angle_measurement	Draw one obtuse angle of 135°.	15	angle
angle_016	clean_scene	angle	angle_measurement	Draw one obtuse angle of 150°.	16	angle
angle_017	clean_scene	angle	angle_measurement	Draw one straight angle (180°). Show it as a straight horizontal line with a semicircular arc above it.	17	angle
angle_018	clean_scene	angle	angle_relations	Draw two adjacent angles of 30° and 60°. The two outer rays are perpendicular.	21	angle
angle_019	clean_scene	angle	angle_relations	Draw two adjacent angles of 45° and 45°. The two outer rays form a right angle.	22	angle
angle_020	clean_scene	angle	angle_relations	Draw two adjacent angles of 20° and 70°. The two outer rays form a right angle.	23	angle
angle_021	clean_scene	angle	angle_relations	Draw two adjacent angles of 55° and 35°.	24	angle
angle_022	clean_scene	angle	angle_relations	Draw two adjacent angles of 72° and 18°.	25	angle
angle_023	clean_scene	angle	angle_relations	Draw two adjacent angles of 15° and 75°.	26	angle
angle_024	clean_scene	angle	angle_relations	Draw two adjacent angles of 40° and 50°.	27	angle
angle_025	clean_scene	angle	angle_relations	Draw two adjacent angles of 33° and 57°.	28	angle
angle_026	clean_scene	angle	angle_relations	Draw two adjacent angles of 25° and 65°.	29	angle
angle_027	clean_scene	angle	angle_relations	Draw two adjacent angles of 48° and 42°.	30	angle
angle_028	clean_scene	angle	angle_relations	Draw two adjacent angles on a straight line: the first is 70° and the second is 110°.	31	angle
angle_029	clean_scene	angle	angle_relations	Draw two adjacent angles on a straight line: the first is 45° and the second is 135°.	32	angle
angle_030	clean_scene	angle	angle_relations	Draw two adjacent angles on a straight line: the first is 60° and the second is 120°.	33	angle
angle_031	clean_scene	angle	angle_relations	Draw two adjacent angles on a straight line: the first is 30° and the second is 150°.	34	angle
angle_032	clean_scene	angle	angle_relations	Draw two adjacent angles on a straight line: the first is 80° and the second is 100°.	35	angle
angle_033	clean_scene	angle	angle_relations	Draw two adjacent angles on a straight line of 90° and 90°. The middle ray is perpendicular to the line.	36	angle
angle_034	clean_scene	angle	angle_relations	Draw two adjacent angles on a straight line: the first is 55° and the second is 125°.	37	angle
angle_035	clean_scene	angle	angle_construction	Draw four rays from one vertex. The four consecutive angles are 90°, 90°, 90°, and 90°. Mark each right angle with the square symbol.	61	angle
angle_036	clean_scene	angle	angle_construction	Draw four rays from one vertex. The four consecutive angles are 30°, 60°, 90°, and 180°.	62	angle
angle_037	clean_scene	angle	angle_construction	Draw four rays from one vertex. The four consecutive angles are 45°, 45°, 135°, and 135°.	63	angle
angle_038	clean_scene	angle	angle_construction	Draw four rays from one vertex. The four consecutive angles are 50°, 70°, 100°, and 140°.	64	angle
angle_039	clean_scene	angle	angle_construction	Draw four rays from one vertex. The four consecutive angles are 80°, 100°, 80°, and 100°.	65	angle
angle_040	clean_scene	angle	angle_construction	Draw four rays from one vertex. The four consecutive angles are 60°, 90°, 120°, and 90°.	66	angle
angle_041	clean_scene	angle	angle_construction	Draw four rays from one vertex. The four consecutive angles are 40°, 50°, 110°, and 160°.	67	angle
angle_042	clean_scene	angle	angle_construction	Draw four rays from one vertex. The four consecutive angles are 75°, 75°, 105°, and 105°.	68	angle
angle_043	clean_scene	angle	angle_construction	Draw four rays from one vertex. The four consecutive angles are 30°, 60°, 120°, and 150°.	69	angle
angle_044	clean_scene	angle	angle_construction	Draw four rays from one vertex. The four consecutive angles are 55°, 65°, 115°, and 125°.	70	angle
angle_045	clean_scene	angle	angle_construction	Draw four rays from one vertex. The four consecutive angles alternate between 45° and 135°: 45°, 135°, 45°, 135°.	71	angle
angle_046	clean_scene	angle	angle_construction	Draw four rays from one vertex. The four consecutive angles are 70°, 50°, 70°, and 170°.	72	angle
angle_047	clean_scene	angle	angle_construction	Draw four rays from one vertex. The four consecutive angles are 36°, 72°, 108°, and 144°.	73	angle
angle_048	clean_scene	angle	angle_construction	Draw four rays from one vertex. The four consecutive angles are 20°, 40°, 60°, and 240°.	74	angle
angle_049	clean_scene	angle	angle_construction	Draw four rays from one vertex. The four consecutive angles are 90°, 60°, 90°, and 120°.	75	angle
angle_050	clean_scene	angle	angle_construction	Draw four rays from one vertex. The four consecutive angles are 60°, 80°, 100°, and 120°.	76	angle
counting_001	clean_scene	counting	exact_counting	An image containing exactly 17 oranges.	13	counting
counting_002	clean_scene	counting	exact_counting	An image containing exactly 18 oranges.	14	counting
counting_003	clean_scene	counting	exact_counting	An image containing exactly 19 oranges.	15	counting
counting_004	clean_scene	counting	exact_counting	An image containing exactly 20 oranges.	16	counting
counting_005	clean_scene	counting	exact_counting	An image containing exactly 21 oranges.	17	counting
counting_006	clean_scene	counting	exact_counting	An image containing exactly 22 oranges.	18	counting
counting_007	clean_scene	counting	exact_counting	An image containing exactly 23 broccoli.	19	counting
counting_008	clean_scene	counting	exact_counting	An image containing exactly 24 broccoli.	20	counting
counting_009	clean_scene	counting	exact_counting	An image containing exactly 25 broccoli.	21	counting
counting_010	clean_scene	counting	exact_counting	An image containing exactly 6 broccoli.	22	counting
counting_011	clean_scene	counting	exact_counting	An image containing exactly 5 broccoli.	23	counting
counting_012	clean_scene	counting	exact_counting	An image containing exactly 8 carrots.	24	counting
counting_013	clean_scene	counting	exact_counting	An image containing exactly 7 carrots.	25	counting
counting_014	clean_scene	counting	exact_counting	An image containing exactly 10 carrots.	26	counting
counting_015	clean_scene	counting	exact_counting	An image containing exactly 9 carrots.	27	counting
counting_016	clean_scene	counting	exact_counting	An image containing exactly 12 carrots.	28	counting
counting_017	clean_scene	counting	exact_counting	An image containing exactly 11 carrots.	29	counting
counting_018	clean_scene	counting	exact_counting	An image containing exactly 14 carrots.	30	counting
counting_019	clean_scene	counting	exact_counting	An image containing exactly 20 oranges.	38	counting
counting_020	clean_scene	counting	exact_counting	An image containing exactly 22 broccoli.	39	counting
counting_021	clean_scene	counting	exact_counting	An image containing exactly 24 carrots.	40	counting
counting_022	clean_scene	counting	exact_counting	An image containing exactly 30 oranges.	43	counting
counting_023	clean_scene	counting	exact_counting	An image containing exactly 7 broccoli.	44	counting
counting_024	clean_scene	counting	exact_counting	An image containing exactly 9 carrots.	45	counting
counting_025	clean_scene	counting	exact_counting	A pile of broccoli on a table. There are exactly 18 broccoli in the pile.	48	counting
counting_026	clean_scene	counting	attribute_based_counting	An image containing exactly 5 red apples.	1	counting
counting_027	clean_scene	counting	attribute_based_counting	An image containing exactly 6 red apples.	2	counting
counting_028	clean_scene	counting	attribute_based_counting	An image containing exactly 7 red apples.	3	counting
counting_029	clean_scene	counting	attribute_based_counting	An image containing exactly 8 red apples.	4	counting
counting_030	clean_scene	counting	attribute_based_counting	An image containing exactly 9 red apples.	5	counting
counting_031	clean_scene	counting	attribute_based_counting	An image containing exactly 10 red apples.	6	counting
counting_032	clean_scene	counting	attribute_based_counting	An image containing exactly 11 green apples.	7	counting
counting_033	clean_scene	counting	attribute_based_counting	An image containing exactly 12 green apples.	8	counting
counting_034	clean_scene	counting	attribute_based_counting	An image containing exactly 13 green apples.	9	counting
counting_035	clean_scene	counting	attribute_based_counting	An image containing exactly 14 green apples.	10	counting
counting_036	clean_scene	counting	attribute_based_counting	An image containing exactly 15 green apples.	11	counting
counting_037	clean_scene	counting	attribute_based_counting	An image containing exactly 16 green apples.	12	counting
counting_038	clean_scene	counting	attribute_based_counting	A realistic orchard scene with three apple trees at different depths. Count only the fully visible red apples hanging on branches of the center tree: there must be exactly 17. Other trees may contain apples of non-red colors only. No baskets, no fallen fruit.	31	counting
counting_039	clean_scene	counting	attribute_based_counting	A realistic top-down photo of a wooden table containing many objects: red apples, red tomatoes, and red balls. Include exactly 19 red apples and at least 10 non-apple red objects to create distractors. Apples must remain clearly identifiable by stems and natural texture.	32	counting
counting_040	clean_scene	counting	attribute_based_counting	A realistic market stall scene with apples under mixed warm-cool lighting and partial occlusions by leaves and price tags. Ensure exactly 21 red apples are visible (partially visible counts if clearly identifiable as apples). Avoid any mirrored reflections that duplicate apples.	33	counting
counting_041	clean_scene	counting	attribute_based_counting	A realistic apple tree in light rain with water droplets and dense foliage. Exactly 23 red apples should be present, with varied sizes due to perspective, but each apple boundary should remain separable for counting. No additional red fruits.	34	counting
counting_042	clean_scene	counting	attribute_based_counting	A realistic harvest scene with one large apple tree and complex branch overlap. Place exactly 25 red apples across near, mid, and far branches. Introduce mild motion blur on leaves only, while apples stay sharp enough to count unambiguously.	35	counting
counting_043	clean_scene	counting	attribute_based_counting	An image containing exactly 16 red apples.	36	counting
counting_044	clean_scene	counting	attribute_based_counting	An image containing exactly 18 green apples.	37	counting
counting_045	clean_scene	counting	attribute_based_counting	An image containing exactly 26 red apples.	41	counting
counting_046	clean_scene	counting	attribute_based_counting	An image containing exactly 28 green apples.	42	counting
counting_047	clean_scene	counting	attribute_based_counting	A basket containing red apples and green apples. There are exactly 13 red apples.	46	counting
counting_048	clean_scene	counting	attribute_based_counting	A market stand with oranges stacked in a display. Exactly 15 oranges are visible.	47	counting
counting_049	clean_scene	counting	attribute_based_counting	A fruit display with oranges arranged in rows. Exactly 24 oranges are present.	50	counting
counting_050	clean_scene	counting	attribute_based_counting	An image with red apples and green apples. There are exactly 6 red apples and 9 green apples. Count only the red ones.	51	counting

End of preview. Expand in Data Studio

MathGen

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

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}
}

Downloads last month: 470

Paper for arielyao/MathGen

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

Paper • 2603.27959 • Published Mar 30