Anagh Malik

I am a fourth-year Computer Science PhD student at the University of Toronto, supervised by David Lindell. I am affiliated with both the Toronto Computational Imaging Group and the Vector Institute. My work has been previously recognized with the CVPR 2025 Best Student Paper Award.

Previously, I completed my MRes at the Dyson Robotics Lab at Imperial College London, where I worked on self-supervised segmentation, under the supervision of Andrew Davison and Ronald Clark.

Email:
anagh [at] cs [dot] toronto [dot] edu


Twitter  /  Resume /  Google Scholar /  GitHub

I am broadly interested in scene understanding. That is using visual cues to infer properites of objects and scenes.

A framework for learning compact latent representations of dynamic 4D objects from unstructured point clouds, enabling video-to-4D generation, cloth simulation, and 3D point tracking.

Recovers camera pose and depth maps from noisy low-light raw images for dense neural reconstruction, enabling accurate 3D scene recovery in extremely low-light conditions.

Time-resolved relighting and geometry estimation through radiance caching.

We propose a passive single‑photon method that opportunistically recovers time‑of‑flight from ambient pulsed light sources.

Novel view synthesis of arbitrary light propagation videos - including effects such as scattering or interreflections.

We introduce a method to do novel view lidar synthesis, allowing sparse view scene reconstruction.

We develop a method for self-supervised segmentation through agreement and self-distillation.

We train an encoder-decoder network to map from xy-coordinates to RGB values and semantic classes, allowing real-time segmentation of an image.

Template stolen from Jon Barron's website.