A Multi-Attribute Dataset of Commercial Bottles

This repository contains a curated dataset of 5,431 high-resolution images of commercial beer bottles, designed specifically for computer vision and machine learning research.

Each image represents a controlled configuration of bottle attributes, including:

bottle type

glass color

fill level

liquid color

label presence

cap state

The dataset utilizes a deterministic file naming convention that directly encodes these attributes, making it highly structured and easy to parse.

In addition to the raw images, the dataset is accompanied by machine-readable JSON annotations and a suite of Python utility scripts for data curation, validation, and statistical analysis. This ensures the integrity of the dataset and provides researchers with ready-to-use tools for their own image processing, object detection, or classification pipelines.

Naming Convention

File naming follows the pattern:

{type}_{color}_{fill}_{liquid}_{label}_{cap}_{index}.{extension} 

For example:

vichy_brown_filled_light_labeled_crowned_001.jpg

Each component corresponds to one categorical attribute of the bottle.