jonathan-roberts1/GRAB-lite

GRAB: A Challenging GRaph Analysis Benchmark for Large Multimodal Models

GRAB consists of 3 splits: GRAB, GRAB-real and GRAB-lite. This is the dataset for GRAB-lite.

Dataset Summary

The 'lite' split of the GRAB benchmark is comprised of 500 questions from GRAB and GRAB-lite. This split is intended as a lightweight evaluation, more suitable for "thinking" models.

Large multimodal models (LMMs) have exhibited proficiencies across many visual tasks. Although numerous benchmarks exist to evaluate model performance, they increasingly have insufficient headroom and are unfit to evaluate the next generation of frontier LMMs.

To overcome this, we present GRAB, a challenging benchmark focused on the tasks human analysts might typically perform when interpreting figures. Such tasks include estimating the mean, intercepts or correlations of functions and data series and performing transforms.

We evaluate a suite of 20 LMMs on GRAB, finding it to be a challenging benchmark, with the current best model scoring just 21.0%.

Example usage

from datasets import load_dataset

# load dataset
grab_lite_dataset = load_dataset("jonathan-roberts1/GRAB-lite", split='GRAB_lite')
"""
Dataset({
    features: ['pid', 'question', 'answer', 'decoded_image', 'task', 'category', 'complexity', 'image'],
    num_rows: 500
})
"""
# query individual questions
grab_lite_dataset[40] # e.g., the 41st element
"""
{'pid': 'grab_1818', 'question': 'Estimate the mean domain length of the functions shown in the plot. Give your answer as an integer.',
'answer': '9', 'decoded_image': <PIL.PngImagePlugin.PngImageFile image mode=RGBA size=2281x2183 at 0x1320FA560>,
'task': 'function', 'category': 'Range', 'complexity': 1, 'image': 'images/grab/1818.jpg'}
"""
question_40 = grab_lite_dataset[40]['question'] # question
answer_40 = grab_lite_dataset[40]['answer'] # ground truth answer
pil_image_40 = grab_lite_dataset[0]['decoded_image']

Note -- the 'image' feature corresponds to filepaths in the images dir in this repository: (https://huggingface.co/datasets/jonathan-roberts1/GRAB-lite/tree/main/images.zip)

Please visit our GitHub repository for example inference code.

Dataset Curators

This dataset was curated by Jonathan Roberts, Kai Han, and Samuel Albanie

Citation Information

@inproceedings{roberts2025grab,
  title={GRAB: A challenging graph analysis benchmark for large multimodal models},
  author={Roberts, Jonathan and Han, Kai and Albanie, Samuel},
  booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision},
  pages={1644--1654},
  year={2025}
}