Susu Mi: A body measurement estimation for African women

Abstract

The issue of finding the “right clothing size” is always a struggle for online shoppers. For this reason, there is a digital gulf between users and e-commerce platforms during buying scenarios. Though the emergence of e-commerce has revolutionized the fashion industry, the issue of clothing fit consistently drives significant market back into traditional shops. With the growing demands of consumers for products tailored to their individual needs and style, there is an increase in demand for custom fitted and custom designed garments. There is also a need for local designers to commercialize their craft in custom-fitted clothing without the bridge of receiving inaccurate measurements. For this gap, there is a need for a cost-effective non-contact body measurement system. In this capstone project report, we explore the use of Convolutional Neural Networks and LSTM model to extract seventeen body measurements from 2d images automatically (front, back, and side). After training on 10 and 50 epochs, the validation loss decreases with each iteration. This means, there is more room for the model to learn. Limitations in dataset size caused a significant impact in the general accuracy of the model. However, in future work, this issue would be solved to ensure higher efficiency.