Sites Inria

Version française

Research team seminary

Learning to Estimate Illumination from a Single Image

We propose an automatic method to infer high dynamic range illumination from a single, limited field-of-view, low dynamic range photograph of a scene. 

  • Date : 10/10/2018
  • Place : Inria de Paris - 2 rue Simone Iff 75012 - Bâtiment C - Salle Lions
  • Guest(s) : Jean - François Lalonde ( Université de Laval, Québec, Canada)

In contrast to previous work that relies on specialized image capture, user input, and/or simple scene models, we frame this as a learning problem and train an end-to-end deep neural network that directly regresses a limited field-of-view photo to HDR illumination. Our algorithm does not rely on strong assumptions on scene geometry, material properties, or lighting. This allows us to automatically recover high-quality HDR illumination estimates that significantly outperform previous state-of-the-art methods. Consequently, using our illumination estimates for applications like 3D object insertion achieves photo-realistic results.

Keywords: Inria Paris Seminary Jean - François Lalonde Research team Laval University