metadata

task_categories:
  - image-text-to-text
license: other
language:
  - ru
tags:
  - mathematics
  - education
  - vlm
  - assessment
  - russian
  - handwritten

EGE Math Solutions Assessment Benchmark

Paper | Code

Dataset Description

This dataset contains student solutions to Russian Unified State Exam (EGE) mathematics problems, with reference scores for benchmarking automated evaluation systems.

🖼️ All images are properly embedded and accessible!

The dataset includes three types of images for each solution:

Student solutions with correct answers shown (152 images)
Student solutions without answers for blind evaluation (152 images)
True/reference solutions for each problem (144 images)

Dataset Statistics

Total examples: 122
Total images: 448
Task types: 7
Score range: 0-4 points

Task Types

Task Type	Count
Financial mathematics	15
Logarithmic inequalities	19
Number theory problem	16
Planimetric problem	17
Problem with parameters	16
Stereometric problem	18
Trigonometric equations	21

Score Distribution

Score	Count	Percentage
0	28	23.0%
1	40	32.8%
2	35	28.7%
3	11	9.0%
4	8	6.6%

Dataset Structure

Each example contains:

solution_id: Unique identifier for the solution
task_id: Task type ID (13-19)
example_id: Specific example identifier
task_type: Description of the task type in English
score: Reference score (0-4)
parts_count: Number of parts in the solution
images_with_answer: List of PIL Images containing student solution with correct answer
images_without_answer: List of PIL Images containing only student solution
images_with_true_solution: List of PIL Images containing task with true solution

Usage

from datasets import load_dataset

# Load the dataset
dataset = load_dataset('Karifannaa/EGE_Math_Solutions_Assessment_Benchmark')

# Access an example
example = dataset['train'][0]
print(f"Solution ID: {example['solution_id']}")
print(f"Task Type: {example['task_type']}")
print(f"Score: {example['score']}")

# View images (all images are PIL Image objects)
print(f"Images with answer: {len(example['images_with_answer'])}")
print(f"Images without answer: {len(example['images_without_answer'])}")
print(f"Images with true solution: {len(example['images_with_true_solution'])}")

# Display an image
if example['images_with_answer']:
    img = example['images_with_answer'][0]
    img.show()  # This will work - images are properly embedded!

Image Access

All images are stored as PIL Image objects and can be directly accessed:

# Get first example
example = dataset['train'][0]

# Access different types of images
student_solution_with_answer = example['images_with_answer'][0]
student_solution_without_answer = example['images_without_answer'][0] 
true_solution = example['images_with_true_solution'][0]

# Images are PIL Image objects with standard methods
print(f"Image size: {student_solution_with_answer.size}")
print(f"Image mode: {student_solution_with_answer.mode}")

# Save image
student_solution_with_answer.save("solution.png")

License

This dataset is provided for research and educational purposes.

Citation

If you use this work in your research, please consider citing it.

Plain Text:

Khrulev, R. (2025). CHECK-MAT: Checking Hand-Written Mathematical Answers for the Russian Unified State Exam. arXiv preprint arXiv:2507.22958. https://arxiv.org/abs/2507.22958

BibTeX:

@misc{khrulev2025checkmatcheckinghandwrittenmathematical,
      title={CHECK-MAT: Checking Hand-Written Mathematical Answers for the Russian Unified State Exam}, 
      author={Ruslan Khrulev},
      year={2025},
      eprint={2507.22958},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2507.22958}, 
}