Rendering-Aware HDR Environment Map Prediction from a Single Image

High dynamic range (HDR) illumination estimation from a single low (LDR) image is significant task in computer vision, graphics, and augmented reality. We present two-stage deep learning-based method to predict an HDR environment map narrow field-of-view LDR image. first learn hybrid parametric representation that sufficiently covers high- low-frequency components the environment. Taking estimated illuminations as guidance, we build generative adversarial network synthesize enables realistic rendering effects. specifically consider effect by supervising networks using losses both stages, on predicted well representation. Quantitative qualitative experiments demonstrate our approach achieves lower relighting errors for virtual object insertion preferred users compared state-of-the-art methods.