JCNF (Joint Convolutional Neural Fields) Model

Version 1.0 (19 Feb. 2019)

Contributed by Seungryong Kim (srkim89@yonsei.ac.kr).

This code is written in MATLAB, and implements the JCNF descriptor [website].

Dependencies

• Download [VLFeat] and [MatConvNet].
• Download the datasets:
  • [MPI-SINTEL Benchmark] ([external link]);

Getting started

• Global_Depth_Network: scripts for learning global depth network
• Depth_Intrinsic_Gradient_Network: scripts for learning depth/intrinsic gradient network
• Gradient_Scale_Network: scripts for learning depth/intrinsic-scale network
• Inference: scripts for inference

Pre-trained Models

• Global Depth Model: [depth-net.mat].
• Depth Intrinsic Gradient Model: [depth-intrinsic-gradient.mat].
• Gradient Scale Models: [depth-gradient-scale.mat], [albedo-gradient-scale.mat], [shading-gradient-scale.mat].

Inference

• Inference/main.m: predicts depth, albedo, and shading simultaneously

Notes

• The code is provided for academic use only. Use of the code in any commercial or industrial related activities is prohibited.
• If you use our code, please cite the paper.

@InProceedings{kim2016,
author = {Seungryong Kim and Kihong Park and Kwanghoon Sohn and Stephen Lin},
title = {Unified Depth Prediction and Intrinsic Image Decomposition from a Single Image via Joint Convolutional Neural Fields},
booktitle = {European Conference on Computer Vision (ECCV)},
year = {2016}
}