Publications
HumanOLAT: A Large-Scale Dataset for Full-Body Human Relighting and Novel-View Synthesis
International Conference for Computer Vision (ICCV) 2025Pulkit Gera, Timo Teufel, Xilong Zhou, Umar Iqbal, Pramod Rao, Jan Kautz, Vladislav Golyanik, Christian Theobalt
We present HumanOLAT, the first public dataset with full-body humans captured in a custom-built lightstage using 331 programmable lights and 40 synchronized cameras. This setup enables high-resolution multi-view OLAT captures under diverse HDR lighting conditions, including white lights, color gradients, and environment maps. HumanOLAT sets a new standard for benchmarking relighting and novel-view synthesis on complex, real human appearances.
Casual Indoor HDR Radiance Capture from Omnidirectional Images
British Machnine Vision Conference (BMVC) 2022Pulkit Gera, Mohammed Reza Karimi Dastjerdi,Charles Renaud, P.J Narayanan, Jean-François Lalonde
We present PanoHDR-NeRF, a novel pipeline to casually capture a plausible full HDR radiance field of a large indoor scene without any elaborate setups or complex capture protocols. AN LDR2HDR network is trained to recover HDR light probes from casually captured omnidirectional LDR frames. It is then used to train PanoHDR-NeRF. Our pipeline can estimate full HDR panoramas from any location of the scene and can synthesize correct lighting effects which enable augmentation of indoor scenes with synthetic objects.
Neural View Synthesis and Appearance Editing from Unstructured Images
Indian Conference on Computer Vision Graphics and Image Processing (ICVGIP) 2021Pulkit Gera, Aakash.KT, Dhawal Sirikonda, P.J Narayanan
We present a neural rendering framework for simultaneous view synthesis and appearance editing of a scene with known environmental illumination captured using a mobile camera. Our approach explicitly disentangles the appearance and learns a lighting representation that is independent of it. We show results of editing the appearance of real scenes in interesting and non-trivial ways.
