GratNet: A Photorealistic Neural Shader for Diffractive Surfaces
N. Kandel, D. S. Dhillon · CEIG, 2026
Research
Computer graphics × machine learning
I am a Ph.D. candidate in Computer Science at Clemson University, working at the intersection of computer graphics, machine learning, and physically based rendering. My research focuses on implicit neural representations for modeling complex surface appearance, particularly diffractive materials.
Machine Learning
Computer Graphics
Neural Rendering
Physically Based Rendering
Implicit Neural Representations
High-Performance Computing
Current work
My current work centers on neural representations for physically based appearance modeling. My selected project, GratNet, develops a lightweight implicit neural representation for rendering diffractive surfaces by combining deep learning with wave-optics-based modeling. The approach reduces memory requirements by up to 50× compared with lookup-table representations.
Background
Before my doctoral studies, I worked in software engineering on web applications, data-driven systems, backend platforms, and location-based products.
Education
Clemson University · Clemson, SC
Ph.D. in Computer Science, expected 2026 · M.S. in Computer Science, 2025
Tribhuvan University, Pulchowk Campus · Lalitpur, Nepal
M.S. in Computer Systems & Knowledge Engineering, 2016 · B.E. in Computer Engineering, 2013
Experience
Graduate Research Assistant · Clemson University · 2021–Present
Research in machine learning, neural representations, and physically based appearance modeling.
Graduate Teaching Assistant · Clemson University · 2024–Present
Teaching and student support across computer graphics, software engineering, programming, and related computing courses.
Machine Learning Intern · LexisNexis Risk Solutions · 2023
Research on scalable distributed neural-network training using TensorFlow, Keras, and HPCC Systems.
Independent Software Engineer · Freelance · 2016–2020
Delivered software engineering projects for multiple clients, completing more than 5,500 hours of freelance work.
Founding Software Engineer · nLocate Pvt. Ltd. · 2014–2016
Developed backend, data-processing, and web systems for location-based search and related products.
Associate Software Engineer · Deerwalk Services Pvt. Ltd. · 2013–2014
Contributed to healthcare analytics software and backend systems.
Skills
Graphics & ML: Neural Rendering, Physically Based Rendering, Deep Learning, Machine Learning, Wave Optics
Programming: Python, C++, MATLAB, Java, JavaScript
Frameworks: PyTorch, TensorFlow, Keras, Django
Computing & Data: GPU Computing, Apache Spark, PostgreSQL, BigQuery, Slurm, Git
Publications
Selected Publications
A selection of publications and research work from my CV.
Improving Performance and Accuracy of Distributed Neural Networks with Generalized Neural Network
N. Kandel, L. Xu, R. Dev · Future of Information and Communication Conference, 2025
Selected project
GratNet: Neural Representation for Diffractive Surfaces · 2022–Present
A lightweight implicit neural representation for physically based rendering of diffractive surfaces.
Technologies: Python, PyTorch, C++, MATLAB, GPU Computing.
About
Narayan Kandel
Ph.D. candidate in Computer Science · Researcher · Software Engineer
Honors
Outstanding Graduate Teaching Assistant, School of Computing, Clemson University (2025).
Nepal Bidya Bhusan Padak (Class B), Presidential Academic Honor awarded for ranking first in the M.S. Computer Systems & Knowledge Engineering program (2018)
Service & leadership
CECAS Graduate Student Advisory Board, Clemson University · 2022–Present
President (2024–2025), Secretary (2023–2024), and School of Computing representative.
ACM SIGGRAPH Student Volunteer · 2023