SmokeSeer

CMU · 2023 – 2025

MS Thesis · 3D Gaussian splatting for wildfire scenes · Project page · 3DV 2026

  • Decomposes wildfire scenes into smoke and surface Gaussians, rendering smoke-free views to assist firefighting.
  • Integrates fluid particle hydrodynamics into the splatting pipeline to model the temporal behavior of smoke.
  • Adapts Mast3r-SfM to localize RGB and thermal cameras in a shared coordinate frame.

Needle: Deep Learning Framework

CMU · Fall 2024

Deep Learning Systems course · Report

  • Differentiable-programming library from scratch with C and CUDA backends, gradient accumulation, and NCCL/Gloo distributed training.
  • Trained a Llama-3 architecture on OpenWebText across 8×4090 nodes with the custom backend.

Multiview Diffusion Models

CMU · Spring 2024

Learning for 3D Vision course · Multiview image-to-image translation

  • Multiview-aware diffusion model for domain translation (e.g. day-to-night) with a conditional U-Net aggregating features along epipolar constraints.

RGB-Thermal Sensor Fusion

CMU / IIT Madras · 2022 – 2023

ADAS research · Advisors: Prof. Aswin Sankaranarayanan, Prof. Kaushik Mitra

  • Physics-based capture of aligned low-light RGB, well-lit RGB, and thermal images of the same scene.
  • Cross-attention architecture that super-resolves thermal images using RGB guidance.