I am Nitthilan, pursuing PhD in AI/ML for Video/Image/3D Graphics. Interested in image and video based motion capture using pose based estimation (SMPLX, FrankMocap, SMPLify-X), clothing and hair 3D model representation using Implicit Neural Representation (NeRF, NSVF, Neural Body) and Deep generative models (GAN, VQVAE) .

• 12 years of software product development experience with wide exposure in building end to end systems from embedded to web technologies.
• Passionate about developing optimized algorithms and exploring different domains. Developed video applications for conferencing, broadcast, and storage. Worked on ME/RC for H264/MPEG and developed plugin/webRTC/Html5 based video endpoints

Current Focus Area:

• Extracting facial pose and full body motion capture (dance steps, sport stances, exercise and fitness poses) data from monocular video (YouTube, TikTok) using SMPLX prior (FrankMocap, SMPLify-X, VIBE)
• Controllable cloth, hair attribute and facial mocap of dynamic humans using SMPLX anchored latent codes (NeuralBody, FrankMocap, HumanParser)
• Accelerating learning of the 3D Neural Scene representation using smaller neural nets and latent representation at a voxel level [NeRF, NSVF, KiloNeRF].

PhD Research: My research is at the intersection of Machine Learning (ML) and Computing Systems (Sys). The overarching theme of my research is to bridge these two areas. Specifically, I’m working towards the vision of Edge AI to efficiently deploy AI solutions for emerging applications (e.g., robotics, selfdriving cars, augmented/virtual reality, and smart health) on edge platforms that are constrained by resources (power, compute, and memory).

(a) Developed a novel hardware-aware design and optimization framework to trade-off power, performance, and accuracy for performing inference using deep neural networks.

(b) Studied effective instantiations of this framework for different applications - Convolutional neural networks for image classification, Graph convolutional networks for 3D computer vision, and Generative adversarial networks for image manipulation tasks.

(c) Published papers at top-tier venues including three journal papers (ACM/IEEE Transactions on CAD and Embedded Computing Systems) and four conference papers (AAAI, DAC, DATE, ICCAD). One journal paper is under review at Journal of Artificial Intelligence Research.