Dataset Card for Sneakers Dataset

Dataset Details

Dataset Description

This dataset contains approximately 93,000 images of sneakers labeled with the manufacturer and model. The images are scraped from Bing Image Search, while the labels (manufacturer and model) are sourced from Sneakers123, an online sneaker database. The dataset is intended for tasks such as image classification, feature extraction, and potentially for applications in fashion and product recognition.

• Curated by: John Guinness
• Language(s): English
• License: MIT License

Dataset Sources

• Image Source: Bing Image Search
• Label Source: Sneakers123

Uses

Direct Use

This dataset can be used for:

• Training machine learning models for sneaker recognition based on manufacturer and model.
• Feature extraction for vector-based searches or retrieval tasks in fashion-related applications.
• Image classification for sneaker identification tasks.

Out-of-Scope Use

The dataset should not be used for tasks outside of fashion-related image recognition, such as general object detection or unrelated computer vision tasks, without significant adjustments to the data.

Dataset Structure

• Images: approximately 93,000 images of sneakers in various lighting and background conditions.
• Labels: Manufacturer and model information extracted from Sneakers123.

No official train-test splits are provided; users can create their own based on their specific use cases.

Dataset Creation

Curation Rationale

The dataset was created to assist in building and training machine learning models for the classification of sneakers, including their make and model. This can aid in developing apps for sneaker enthusiasts, retailers, and online shopping platforms.

Source Data

Data Collection and Processing

• Data Collection: The images were collected by scraping Bing Image Search for sneaker images, while the corresponding labels were sourced from Sneakers123.
• Processing: Images are stored in their original resolution, and labels are text-based metadata associated with each image.

Who are the source data producers?

• Images: Publicly available sneaker images collected through Bing Image Search.
• Labels: Manufacturer and model names curated from Sneakers123.

Annotations

The dataset does not contain manual annotations beyond the labels scraped from Sneakers123.

Bias, Risks, and Limitations

• Bias: As the images were scraped from search engines, the dataset may exhibit biases in terms of popular brands or models, and might not cover less common or newer sneaker releases comprehensively.
• Risks: There is a risk of mislabeling due to the automated nature of scraping. Users should verify the correctness of the labels before deploying the dataset in production applications.

Recommendations

Users should be cautious about potential biases in brand representation and consider performing additional filtering or augmentation if required for their specific use case.

Dataset Card Authors

John Guinness