Saurav Dhakad

I am a Senior Machine Learning Engineer with a Master's in Computer Science from Columbia University. My work focuses on 3D AI, deep generative models, and scalable systems for content creation using diffusion models and multimodal learning. I'm currently a Senior Machine Learning Engineer at Roblox within the Foundation AI organization, where I work on 3D Generative AI and production-grade pipelines for creating and manipulating 3D content at platform scale.

Prior to Roblox, I was on the Avatars R&D team at Genies Inc, developing generative AI for expressive, controllable 3D avatars, spanning diffusion-based synthesis, view-consistent generation, mesh reconstruction, and multimodal conditioning. Before that, I spent several years at Unity Technologies, working on applied ML research and computer vision problems like human pose estimation, saliency detection, and synthetic data generation. I’ve also contributed to open-source perception pipelines and published work on synthetic dataset generation for human-centric vision tasks (Unity Perception, PSP-HDRI+).

My broader research interests include Gen AI, Vision, LLMs, and efficient learning for real-time systems. I'm passionate about building tools that bridge the physical and virtual worlds, and enabling scalable AI-powered experiences.

Publications

OccFace: Unified Occlusion-Aware Facial Landmark Detection with Per-Point Visibility

ICPR 2026 [Paper]

PSP-HDRI+: A Synthetic Dataset Generator for Pre-Training of Human-Centric Computer Vision Models

ICML 2022 [WebPage] [Paper] [Code]

PeopleSansPeople: A Synthetic Data Generator for Human-Centric Computer Vision

arXiv 2021 [Paper] [Code]

Unity Perception: Generate Synthetic Data for Computer Vision

arXiv 2021 [Paper] [Code]

• Google Scholar

Blogs & Talks

• Blog: Training a Performant Object Detection ML Model on Synthetic Data Using Unity
• Podcast: TechGuide Podcast: Applying ML to 3D Animations with Saurav Dhakad

Experience