ai2dcaption

  • Description:

This dataset is primarily based off the AI2D Dataset (see here).

See Section 4.1 of our paper for the AI2D-Caption dataset annotation process.

Split Examples
'auditor_llm_training_examples' 30
'gpt4v' 4,903
'llava_15' 4,902
'planner_llm_training_examples' 30
'test' 75
  • Feature structure:
FeaturesDict({
    'caption': Text(shape=(), dtype=string),
    'entities': Sequence({
        'bounds': BBoxFeature(shape=(4,), dtype=float32),
        'cat': ClassLabel(shape=(), dtype=int64, num_classes=10),
        'from': Text(shape=(), dtype=string),
        'id': Text(shape=(), dtype=string),
        'label': Text(shape=(), dtype=string),
        'to': Text(shape=(), dtype=string),
        'type': ClassLabel(shape=(), dtype=int64, num_classes=5),
    }),
    'image': Image(shape=(None, None, 3), dtype=uint8, description=The image of the diagram.),
    'image_filename': Text(shape=(), dtype=string),
    'layout': ClassLabel(shape=(), dtype=int64, num_classes=7),
    'relationships': Sequence(Text(shape=(), dtype=string)),
    'topic': ClassLabel(shape=(), dtype=int64, num_classes=4),
})
  • Feature documentation:
Feature Class Shape Dtype Description
FeaturesDict
caption Text string
entities Sequence
entities/bounds BBoxFeature (4,) float32
entities/cat ClassLabel int64
entities/from Text string
entities/id Text string
entities/label Text string
entities/to Text string
entities/type ClassLabel int64
image Image (None, None, 3) uint8 The image of the diagram.
image_filename Text string Image filename. e.g. "1337.png"
layout ClassLabel int64
relationships Sequence(Text) (None,) string
topic ClassLabel int64

Visualization

  • Citation:
@inproceedings{Zala2024DiagrammerGPT,
        author = {Abhay Zala and Han Lin and Jaemin Cho and Mohit Bansal},
        title = {DiagrammerGPT: Generating Open-Domain, Open-Platform Diagrams via LLM Planning},
        year = {2024},
        booktitle = {COLM},
}