Dataset Card for GRAB

Dataset Summary

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 21.7%.

Example usage

from datasets import load_dataset

# load dataset
grab_dataset = load_dataset("jonathan-roberts1/GRAB", split='GRAB')
"""
Dataset({
    features: ['pid', 'question', 'decoded_image', 'image', 'answer', 'task', 'category', 'complexity'],
    num_rows: 2170
})
"""
# query individual questions
grab_dataset[40] # e.g., the 41st element
"""
{'pid': 40, 'question': 'What is the value of the y-intercept of the function? Give your answer as an integer.',
'decoded_image': <PIL.PngImagePlugin.PngImageFile image mode=RGBA size=5836x4842 at 0x12288EA60>,
'image': 'images/40.png', 'answer': '1', 'task': 'properties', 'category': 'Intercepts and Gradients',
'complexity': 0}
"""
question_40 = grab_dataset[40]['question'] # question
answer_40 = grab_dataset[40]['answer'] # ground truth answer
pil_image_40 = grab_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/resolve/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

@article{roberts2024grab,
  title={GRAB: A Challenging GRaph Analysis Benchmark for Large Multimodal Models},
  author={Roberts, Jonathan and Han, Kai and Albanie, Samuel},
  journal={arXiv preprint arXiv:2408.11817},
  year={2024}
}