Curl Type and Density Classifier

Author : Anisha Ravishankar

Abstract :The curl-type classifier is essentially a model made using EfficientNet B2. It classifies hair into 4 curl types: 2A-2B, 2C, 3A-3B, 3C. Type 2 represents wavy hair and type 3 represents curly hair. This was made to address the under-representation of curly hair in India. The dataset was manually built and labelled, using images from sources like Pinterest, Reddit, etc. It contains 400 images, 100 images per class to ensure balance (which ensures that there is no bias in the results). Models EfficientNet B0 and B1 were also tested before B2. However, B2 achieved the highest accuracy among the 3, which led to it being chosen. Other parameters like learning rate, weight decay, train-validation-test split were adjusted numerous times before achieving the desirable accuracy. The model finally achieved a validation accuracy of 83.33% and a test accuracy of 81.67%. Further, a simple web-interface was created to make the model easy to use. The web-interface includes not only curl-type classification, but also density classification. Density is decided based on a short, 3-question quiz that the user needs to take. Based on the curl-type and density, the model suggests an easy-to-follow, comprehensive hair-care routine (mainly intended for beginners). This website was deployed using HuggingFace to make it accessible for real-world use.

Keywords :EfficientNet-B2 Based Curl-Type Classifier with Web Tool for Curly Hair Care

Conference Name :International Conference on Advances in Mathematics, Engineering & Technology (ICAMET)

Conference Place Amsterdam, Netherlands

Conference Date 3rd Nov 2025