Self-Supervised Depth Estimation via Focal Stack Reconstruction

This project is in proceeding CVPR2023

  • Researched on depth estimation from defocus clue; Proposed a self-learning framework that estimate depth from sparse focal stack; Design the hardware that can capture the focal stack; Address the problems exist in previous works.
  • Tested the framework on NYUv2 dataset and MobileDFD dataset; Proved the method to be SOTA on depth estimation from defocus clues tasks; Verified the ability of the model to be applied in industry.