CRF-net: Single Image Radiometric Calibration using CNNs

Abstract

In this paper we present CRF-net, a CNN-based solution for estimating the camera response function from a single photograph. We follow the recent trend of using synthetic training data, and generate a large set of training pairs based on a small set of radio-metrically linear images and the DoRF database of camera response functions. The resulting CRF-net estimates the parameters of the EMoR camera response model directly from a single photograph. Experimentally, we show that CRF-net is able to accurately recover the camera response function from a single photograph under a wide range of conditions.