Parth Kashikar is an undergraduate student at BITS Pilani, India, majoring in Physics as well as Electronic and Instrumentation Engineering. Parth’s key mathematical interests and past projects lie in the fields of Generative Modeling, Reinforcement Learning, Computer Vision and Temporal Learning. Parth hopes to strengthen the theory behind AI setups, as well as leverage them for accelerating scientific and industrial growth. He believes that Artificial Intelligence has yet to solve complex issues including causality and creativity, and that further explorations in his fields of interest might provide some answers. Parth’s fascination with AI has led him to participate in the Wolfram Summer School, and he wishes to pursue the same after graduation as well.
Computational Essay: Causality : A Brief Introduction
The goal of the project is to write a function that takes a picture of a person’s face as the input and generates a similar looking avatar (emoji) as the output. A data-driven approach involving face segmentation, clustering from a custom avatar segment space and a final reorganization for the output has been adopted for achieving effective results.
Summary of Results
A mapping function was successfully created to map faces to similar-looking emojis with best-fit semantics except edge cases of certain types of hairstyles and spectacles. The function accepts the image of any arbitrary face and produces the corresponding emoji using a combination of deep learning and heuristic setups. The elements currently used for the generation include hairstyle, hair color, skin color and iris color. An eyebrow tracker has been implemented but is not yet included in the current pipeline.
- Integration of some functions to map hairstyles and jawlines together, varying the hairstyle according to the jawline chosen as well as eyebrow tracker inclusion.
- Spectacle detection and improvement of the hair/face segmentation dataset for inclusion of a wider variety of hair.
- Improvement of the hairstyle-fitting function for a better metric of contour-shape similarity.
- Curation of a dataset with pre-decided anchor points of the graphics to manipulate the required semantic features and clearer coloring.