Generative Image Restoration
Latent-space restoration with StyleGAN and diffusion-based priors for deblurring, denoising and inpainting, including unsupervised and training-free settings.
I am a fourth-year Ph.D. student in Computer Science and Engineering at IIT Gandhinagar, advised by Prof. Shanmuganathan Raman. I lead an independent research line on generative priors for image restoration — latent inversion in StyleGAN and diffusion models, zero-shot and training-free restoration, and 2D/3D Gaussian splatting for hyperspectral imaging.
My work sits deliberately across disciplines. I collaborate with computer scientists, earth scientists and agricultural researchers, and I have built and released datasets that serve all three.
Before IIT Gandhinagar, I was a Project Associate at IIT Ropar on hyperspectral-stereo fusion for industrial volume estimation. I hold an M.Tech (2nd rank) and B.Tech (1st rank) in CSE from Aliah University, Kolkata, and a UAV Pilot Licence for platforms up to 25 kg.
I am open to research positions at the boundary of visual learning, generative modelling and real-world sensing. Best way to reach me: akbar.ali@iitgn.ac.in.
Four connected directions currently shape my work.
Latent-space restoration with StyleGAN and diffusion-based priors for deblurring, denoising and inpainting, including unsupervised and training-free settings.
Extending Gaussian representations beyond RGB to model spatial structure and rich spectral information for hyperspectral reconstruction and super-resolution.
Continuous neural representations for recovering high-resolution hyperspectral signals from limited spatial observations, with emphasis on spectral fidelity.
Physics-aware and low-rank representations that exploit the strong spatial-spectral redundancy of hyperspectral observations for robust restoration.
Generative priors, latent inversion, neural fields, Gaussian splatting, low-rank spectral representations.
Deblurring, denoising, inpainting, super-resolution, blind and zero-shot restoration.
Hyperspectral data, spectral reconstruction, spatial-spectral modelling and remote-sensing workflows.
UAV mission planning, multi-altitude aerial imaging, field campaigns and dataset construction.