Silhouette Detection with Neural Networks and Random Forest Models

The study aims to develop a system that can automatically label the silhouette type of clothing in an image to support the applications of recommending clothing silhouettes based on body shapes. The silhouette identification model includes three parts. Part one is a convolution neural network model trained on the DeepFashion2 dataset. This model can detect the clothing contours of an image. Part one is a neural network model that can detect the body joints to help locate the measurements associated with determining clothing silhouette. Part three is a random forest model linking the extracted measurement with silhouette labels. The developed system can accurately categorize the silhouette of clothing in an image into O, X, H, V, A, and S shapes. This project is collaborated with Dr. Jiayin Li.