Praneet Singh

About Me

I am a Sr. Neural Video Processing Engineer at Dolby Laboratories. I develop learned codecs for images, video and 3D scenes, build video foundation models, and contribute to JVET's next-generation video coding standard.

I received my PhD in Electrical and Computer Engineering from Purdue University in May 2025, advised by Dr. Amy R. Reibman and Dr. Edward J. Delp. My thesis, funded by Ford Motor Company, studied task-aware video compression and quality estimation for practical video analytics systems.

A good codec learns a compact representation of how the world looks and moves, and most video foundation models train and generate in such latent spaces. I want to develop compression methods that help video foundation models and world models handle longer, higher-resolution video.

I am also interested in empirical AI safety. During my PhD, I evaluated how compression affects downstream vision systems and stress-tested face-quality metrics at scale. I want to apply that evaluation experience to finding failures in larger models.

Research Interests

• Learned Image and Video Compression
• Video Foundation Models and World Models
• Generative (Diffusion and Flow-Matching) Compression
• 3D Scene Compression (Gaussian Splatting)
• Task-aware Video Analytics and Quality Estimation
• AI Safety: Model Evaluation and Robustness

Technical Skills

• Architectures: Diffusion, Flow Matching, Mixture of Experts, Transformers, Neural Fields, 3D Gaussian Splats, Video Foundation Models
• Training: Variable-Rate Compression, Rate-Distortion Optimization, Pruning and Quantization
• Evaluation: BD-rate Benchmarking vs. VTM/HM, Task-aware Quality Estimation, Stress Testing at Scale (33M+ pairs), Dataset Curation
• Systems: End-to-end Video Analytics Pipelines, Codec Reference Software (HM, VTM), JVET/AOM Standards; PyTorch, C++

Work Experience

• Dolby Laboratories
• Apple Inc.
• Purdue University
• Indian Institute of Science