Deep Generative Model
We are currently working on a research project to create new hairstyles using the trained generator of StyleGAN2. In this research, we are using feature embedding (of various reference images) techniques optimized for a natural style transfer. Specifically, we are trying to find the method for seamless blending between the naturally generated style image and the original image. We are also developing an efficient hairstyle creation solution through various parametric encoding models for interactive style controls using a simple sketche operation.
We are also conducting research on the improvements of NeRF, a neural network-based scene representation model, specifically for few-shot cases and dynamic scene cases. The researching features will be added to the NeRF plugin module of the rendering engine that our laboratory has been developing. The entire pipeline is motivated by Taichi-nerf, based on the core technique, intant NGP, which allows for fast/interactive NeRF.
programming experience
Python, Pytorch, Tensorflow
