ECCV 2020 Daily - Tuesday

3 Zhe Chen 9 Finally, the two parts are integrated into a differentiable Bayesian inverse rendering framework to estimate the reflectance and illumination. To find out more about Zhe’s work, visit his live spotlight session [#4165] and Q&A today (Tuesday) at 02:00 (UTC +1). For the illumination part, the work introduces a deep illumination prior . This illumination model is very high- dimensional but makes use of the inductive bias of convolutional neural networks towards natural image signals and impedance to noise. The deep illumination prior is used as a proxy to DAILY T u s d a y