Research, Collaborations and Field Leadership

From generative image restoration to hyperspectral sensing, my projects combine representation learning with real-world data and interdisciplinary collaboration.

Current Research Directions

HyperPhys-LRGS

Physics-aware low-rank hyperspectral Gaussian splatting for spectral reconstruction and super-resolution. The direction combines compact spectral bases with differentiable Gaussian representations.

HSIGaussian SplattingLow Rank

Hyperspectral Gaussian Splatting

Exploring Gaussian representations for hyperspectral denoising and reconstruction, including spectral bases and residual restoration strategies.

DenoisingSpectral Basis

Implicit HSI Super-Resolution

Adapting implicit neural representations to continuous hyperspectral reconstruction so that spatial queries can recover spectrally consistent high-resolution signals.

LIIFNeural Field

Generative Restoration

Developing latent-space and training-free approaches for difficult real-world image degradations, with emphasis on robustness and faithful reconstruction.

StyleGANDiffusion

Collaborations & Field Work

3D Geophysical Image Translation using GANs
Feb 2024 — Jan 2025
A year-long collaboration with geoscientists, translating 3D geophysical scans into photorealistic virtual outcrop geology.
What this involved

Interdisciplinary problem formulation, generative modelling and communication between geological interpretation and computer-vision representations.

Smart Farming of Cotton using Aerial Imagery — COT-AD
Jun 2023 — Jan 2024
Designed and ran a six-month acquisition campaign with agricultural researchers, flying weekly missions at 10 m, 15 m and 115 m. I owned the flight schedule, the multi-altitude protocol, and the dataset that became COT-AD.
Why the dataset matters

The repeated multi-altitude observations follow the same field across a full cotton growth cycle, connecting acquisition strategy with downstream visual analysis.

Cotton growth monitoring
Weekly flights tracking one field through a full growth cycle.
Drone and Core Data Acquisition — Earth Science, IIT Gandhinagar
2023 — 2024
Planned and led three multi-day campaigns for aerial and core sample collection. Site names withheld for confidentiality.
Field campaign
Field campaign
3D Vision Summer School — IIIT Bangalore
Jun 2024
Week-long school on deep learning for 3D vision. The Gaussian splatting sessions shaped the direction my hyperspectral work took afterwards.
3D Vision Summer School
3D Vision Summer School
Visvesvaraya PhD Scheme Workshop — IIT Mandi
May 2024
Presented my work to an interdisciplinary audience of scholars from across engineering disciplines.
IIT Mandi workshop
IIT Mandi workshop