Use Conditional Generative Adversarial Network-based techniques to generate human face images with semantic text information as an auxiliary input method.
Security organizations (for example police) often need to rely on skills of sketch artists to draw sketches of an accused person. Such sketch artists due to their human limitations cannot draw numerous sketches or variants of the same sketch at a time and on top of that the sketch artist demands a person to actually describe the facial attributes of the accused while he keeps on sketching the face. Such adversary could be countered with the advent of “Conditional Generative Adversarial Network based” methods where a system (machine learning model) could try to render sketches of face images based on some semantic facial attribute text input. Relevant references https://arxiv.org/pdf/1810.11919.pdf, https://arxiv.org/pdf/1811.12784.pdf
Goal : A system (or machine learning model) that can generate a face image based on the semantic text information about the face
Learning outcomes: Image processing, machine learning, real-world implementation and experiments
Qualifications:
Supervisors:
