LAMDA-Tabular/MMTU

Benchmark Datasets used for "Compositional Condition Question Answering in Tabular Understanding"

This repository contains the benchmark used in "Compositional Condition Question Answering in Tabular Understanding". Specifically, we provide the datasets in two zip files: data.zip and picked.zip.

1. data.zip:

This zip file contains four JSON files, including IE.json, CR.json, CC.json and RC.json, each containing all the samples of one category of QA pairs. Each contains following information:

    {
        "question": "How many cartridge types have a 100 m group (moa) less than 0.5?",
        "answer": "3",
        "image_path": "picked/image_0.jpg"
    },

2. picked.zip:

This zip file contains all the images.