Wax-MNIST is an image dataset created for machine learning. It is similar to the well-known MNIST dataset, which contains images of handwritten digits. However, instead of digits, Wax-MNIST images represent colored patterns found on African fabrics, such as wax, kente, or bogolan. Wax-MNIST was created to encourage cultural diversity in AI projects and raise awareness of the richness of African patterns.
0. Context
African fabric, also known as “pagne” in some regions of Africa, has different names depending on the regions and countries of Africa. For example, in West Africa, it is called “wax cloth,” “African cloth,” or “printed cloth,” while in Central Africa, it is often called “pagne fabric” or “African fabric.” There are also many other names and designations for these fabrics, which are often associated with specific patterns and designs.
However, Wax fabric is not a traditional African fabric like Kente, Bogolan, Faso Dan Fani, or other hand-woven fabrics, as it was introduced to West Africa by Dutch and English traders in the 19th century. However, Wax fabric has become very popular in West Africa and is now considered an iconic African fabric. The vibrant patterns and colors of Wax fabric are often associated with African fashion and culture. Despite its relatively recent origin, Wax fabric has become an important part of African clothing culture and is often used for special occasions and important events.
1. Objective
This dataset was created as part of the WaxClassification project. The aim of this project was to provide African Wax fabric customers with a tool to help them purchase Wax fabrics on markets. Technically, the project is based on machine learning and image classification with a model deployment with Android Studio. Indeed, the application allows users to know the brand, quality, price, and recommendations of the fabric they want to buy from an image of the fabric.
2. Image collection
The images were collected online from various e-commerce sites such as Amazon, Cdiscount, tissusplus.com, ebay, etsy.com, afrikrea.com, …, and through Google search image and Facebook APIs. Labeling was done at two levels. A first automated level during collection from keywords and product descriptions of e-commerce sites. A second level was performed manually by the project team and other collaborators.
3. Image characteristics
The current dataset consists of approximately 1081 images of size 64 x 64 or 28 x 28 in 8 categories of fabrics.
There are undoubtedly many types of Wax fabrics, each with its own patterns, colors, and styles. This data concerns the most well-known ones:
- Vlisco: Vlisco is one of the oldest Wax fabric brands, founded in 1846. Their fabrics are characterized by geometric patterns and bright colors.
- Woodin: Woodin is a Ghanaian Wax fabric brand, known for its contemporary and colorful designs.
- Uniwax: Uniwax is an Ivory Coast-based Wax fabric brand that offers a range of fabrics with traditional and modern patterns.
- Java or Hollandais: Java is a Dutch Wax fabric brand that offers a range of fabrics with floral and abstract patterns.
- GTP (Ghana Textiles Printing): GTP fabrics have bold and vibrant patterns, often inspired by nature and daily life. The colors are often bright and the patterns are sometimes embossed.
- Real Wax: Real Wax fabrics are produced in Europe and Asia, but are often associated with Africa. They have traditional patterns such as checks, diamonds, and stars, and are often made with bright and contrasting colors.
- Hitarget: Hitarget fabrics have vibrant and often bold patterns, with bright and contrasting colors. The patterns are often geometric and inspired by nature.
- Batik: Batik is a fabric dyeing technique that involves applying wax to the fabric to create patterns before dyeing it. Batik fabrics can have abstract or figurative patterns, and are often associated with Indonesia
4. Objectives and possible tasks
The Wax-MNIST database was created to facilitate automatic classification of African Wax fabrics and automatic recognition of patterns and symbols in African Wax fabrics. Here are some possible objectives and tasks for this database:
Pattern recognition: Patterns in African Wax fabrics can have important cultural and symbolic meanings. The Wax-MNIST database can be used to train pattern recognition models to identify and classify different patterns in Wax fabrics.
Fabric classification: The Wax-MNIST database can be used to train fabric classification models to differentiate between different types of Wax fabrics.
Fashion application development: Pattern recognition and fabric classification models trained using the Wax-MNIST database can be used to develop fashion applications, such as clothing design tools or style recommendation applications based on user preferences.
Cultural preservation: African Wax fabrics are an important element of African culture. The Wax-MNIST database can be used to help document and preserve these fabrics by digitizing and categorizing fabric patterns and styles.
Education: The Wax-MNIST database can be used to raise awareness about the different types of African Wax fabrics and their cultural significance. This can help promote cultural diversity and combat cultural marginalization.
7. Similar Projects
African Fabric Images : Image Dataset of African Fabric Patterns and Textiles, is made available on Kaggle by AYOMIKUN SAMUEL. This project contains approximately 1056 unlabeled images.