DGE-CNN-demo

DGE-CNN demo for 2D-to-3D holographic display based on depth gradient extracting module and CNN network.

"This source code is for training DGE-CNN network to recover depth map from single RGB 2D input."

"The output depth map is boundary-enhanced and will be used for layer-based angular spectrum calculation of 3D hologram."

"DGE module is packed in loss_gd in utils.py, to generate boundary-enhanced depth maps, train the network with loss_gd."

Author: Ninghe Liu ([email protected])

Usage

run train.py to train the DGE-CNN network, you can choose the pretrained network we provided for easy convergence.

run inference.py to predict the boundary-enhanced depth map for input 2D image.

run segment.py to conduct pixel-average slicing according to predicted depth map and input 2D image.

The layer-based angular spectrum hologram generation is not available in this repository but may be obtained from the authors upon reasonable request.

Data

Download the NYU Depth Dataset V2 Labelled Dataset : http://horatio.cs.nyu.edu/mit/silberman/nyu_depth_v2/nyu_depth_v2_labeled.mat.

Download the Make3D Dataset: http://make3d.cs.cornell.edu/data.html

Or we have provided our back up NYU Depth Dataset V2 data that goes with our demo.

Back up link (data and source code): https://cloud.tsinghua.edu.cn/d/edf599b1febf4c1a955e/

Reference

N. Liu, Z. Huang, Z. He and L. Cao, "DGE-CNN: 2D-to-3D holographic display based on depth gradient extracting module and CNN network" Already submitted to Optics Express.

All rights reserved.