Visual Media and Data Management Lab, SJTU

Created in November 01, 2018

2018

mage Content Understanding Based on Geometric Structure Shanghai, China

  • Proposed a parallel CNN architecture to recover albedo and shading for single image intrinsic decomposition in parallel.
  • The image reconstruction loss function was shared across the parallel networks to maximize reconstruction consistency.
  • Integrated gradient information into the CNN-based model, leveraging physics-based priors to better handle complex real-world scenarios.