Project Description

The goal of our experiment was to create a solution that generates bitmoji (cartoon avatar) from a face image uploaded by the user. Deep face-scan – extract hair, beard, eyebrows, eyes, nose, lips, face color, and shape, together with different accessories: glasses, earrings, piercings, etc. Based on these features we extract similar shapes and colors from the database using deep content-based image retrieval and assemble the bitmoji.

Technologies:

• OpenCV for classical computer vision routines.
• PyTorch based deep CNN models for face scan (segmentation) and representation (vectorization).
• Faiss for fast shape similarity search.
• Scrapy for data collection.