# Classify Gases from Multimodal Information: Data description

*English translation by SOTA – AI Community of the Japanese original, shared under CC BY-NC-SA 3.0 IGO like the original. Organisers who would like this translation removed can email sota.ai.community@gmail.com.*

*Data tab of the Kaggle competition "[Playground] JOAI Competition 2025". On Kaggle the data are 6,403 files (477.07 MB; png, csv), licensed Attribution-NonCommercial-ShareAlike 3.0 IGO (CC BY-NC-SA 3.0 IGO).*

## Files

*   **train.csv** – training set
*   **test.csv** – evaluation set
*   **sample_submission.csv** – sample submission file
*   **images** – directory containing the image files in PNG format. It contains the images for both train and test.

## Columns

**train.csv**
*   `MQ8` – measurement from a gas sensor.
*   `MQ5` – measurement from a gas sensor.
*   `Caption` – text describing the image. The images provided in this competition are partly cropped, but this text was generated from the images before cropping.
*   `image_path_uuid` – the corresponding image in the `images` folder.
*   `Gas` – ground-truth label.

**test.csv**
*   `MQ8` – measurement from a gas sensor.
*   `MQ5` – measurement from a gas sensor.
*   `Caption` – text describing the image. The images provided in this competition are partly cropped, but this text was generated from the images before cropping.
*   `image_path_uuid` – the corresponding image in the `images` folder.

**sample_submission.csv**
*   `index` – the index in **test.csv**.
*   `Gas` – predicted label.