Mustafa Isik Email and Phone Number

•Optimizing the inference code to achieve top performance using Triton, CUDA, torch.compile, and other tools.•Training/fine-tuning video diffusion models for talking avatars.•Built a video scraper, an evaluation framework and parts of the data processing pipeline for video models.•Worked on 3D gaussian avatars from monocular videos.

•Worked on diffusion models for avatars and environments.•Researched, designed and implemented a volumetric neural rendering framework for 3D digital humans that can speak and perform gestures to communicate.•Wrote highly-optimized CUDA kernels and used Nvidia OptiX for fast ray tracing to have custom forward and backward passes to achieve real-time rates for 3D avatars.

•Devised a compact and efficient method, HumanRF (published at SIGGRAPH 2023), that enables novel-view synthesis for arbitrarily long videos of dynamic 3D humans.•Worked on various tasks to prepare a processing pipeline for multi-view human avatar datasets including a memory-efficient, on-the-fly dataloading mechanism for terabytes of data.•Worked on incorporating hand gestures into digital avatars using machine learning and computer vision techniques.

•Devised a method for neural rendering of human heads from sparse set of views.

•Invented a deep learning-based image-space denoising method for interactive Monte Carlo rendering.•The work is patented, published at SIGGRAPH 2021 and shipped in Adobe products.

•Researched fast volume rendering and visualization techniques.

•Developed an application that collects player statistics in a video game.