LingoQA Datasets README

Overview

The LingoQA datasets comprise a collection of complementary datasets designed for training and evaluating machine learning models on video understanding and question-answering tasks. These datasets are categorized into three main types: action, scenery, and evaluation, each containing video segments, questions, and answers, along with associated images to aid in visual understanding tasks.

Directory Structure

The datasets are organized as follows:

LingoQA/
├── action/
│   ├── train.parquet          # Training data for action-related questions
│   └── images.zip             # Zipped directory of images related to action segments
├── scenery/
│   ├── train.parquet          # Training data for scenery-related questions
│   └── images.zip             # Zipped directory of images related to scenery segments
└── evaluation/
    ├── val.parquet            # Validation data for evaluating models
    └── images.zip             # Zipped directory of images related to evaluation segments

Data Format

Each .parquet file within the datasets contains the following columns:

segment_id: An md5 hash that uniquely identifies each video segment.
question_id: A unique identifier for each question associated with a video segment.
images: An array of relative paths pointing to images that correspond to the video segment. Example format: ['images/train/hash_segment_1/0.jpg', ...].
question: The question text related to the video segment.
answer: The answer text related to the question posed.

How to Use

Preparing the Data

Extract Images: First, unzip the images.zip files in their respective directories to access the images associated with each dataset.

unzip action/images.zip -d action/
unzip scenery/images.zip -d scenery/
unzip evaluation/images.zip -d evaluation/

Load Datasets: You can load the .parquet files using Python with libraries such as pandas or pyarrow. Here's an example of how to load a dataset:

import pandas as pd

# Load an example dataset
action_train = pd.read_parquet('action/train.parquet')

Using the Data

Model Training: Use the train.parquet files from the action and scenery datasets to train your machine learning models. These datasets provide a rich set of questions and answers along with images to facilitate training models capable of understanding and responding to queries about video content.
Model Evaluation: The evaluation/val.parquet file is intended for validating the performance of your models. It offers a separate set of questions and answers to test your model's ability to generalize to new data.

License

Please ensure to review the license agreement associated with the LingoQA datasets before using them for your projects.