peterzs / svbrdf-gan

Joint SVBRDF Recovery and Synthesis

Source code of our paper:

Joint SVBRDF Recovery and Synthesis From a Single Image using an Unsupervised Generative Adversarial Network ** (EGSR 2020)

Yezi Zhao, Beibei Wang*, Yanning Xu, Zheng Zeng, Lu Wang, Nicolas Holzschuch

(* joint first author)

Requirements

• Python 3.6, numpy, Tensorflow-gpu 1.4
• 8G GPU Memory (test on GTX 1070)

Usage

The following command will train the network on a captured image, and output intermediate result during training as well as recovered and synthesized SVBRDF maps of a cental cropped tile of the captured image (considering gpu memory).

python train.py --input_dir aittalaimgs/plastic_red_input.jpg --log_dir logs/0614-plastic_red --output_dir predictions/0614-plastic_red --crop_size 800

--input_dir path to the captured image. We offer several images for testing, and the whole captured images dataset can be find here: https://mediatech.aalto.fi/publications/graphics/TwoShotSVBRDF/

--log_dir folder to save checkpoint.

--output_dir folder to save recovered SVBRDF maps.

--crop_size define the size of a cropped tile from the captured image, default is 800. This tile is to be sent to the network and get 2× size SVBRDF maps. crop_size depends on the GPU memory, larger size might cause OOM error.

--img_h, --img_w define the height and width of your captured image, default 1224* 1632.

Citation

Acknowledgements

Implementation based on https://team.inria.fr/graphdeco/projects/deep-materials/ and modified. We sincerely thank for their great work.