Skip to content

amirhertz/geometric-textures

master
Switch branches/tags

Name already in use

A tag already exists with the provided branch name. Many Git commands accept both tag and branch names, so creating this branch may cause unexpected behavior. Are you sure you want to create this branch?
Code

Latest commit

 

Git stats

Files

Permalink
Failed to load latest commit information.
Type
Name
Latest commit message
Commit time
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Deep Geometric Texture Synthesis in PyTorch

SIGGRAPH 2020 [Paper] [Project Page]

Deep Geometric Texture Synthesis is an approach for learning the local geometric textures present within a 3D mesh model. This can be used to learn the unknown 3D geometric texture statistics from a single 3D model, and then synthesizes them on different 3D models. This repository contains the code for:

(1) creating multi-scale training data

(2) training a series of multi-scale generators

(3) synthesizing the learned geometric textures on unseen models

Installation

  • Clone this repo git clone https://github.com/amirhertz/geometric-textures.git.
  • Install via conda environment conda env create -f environment.yml (creates an environment called dgts)

Jupyter Demo

We provide an End-to-End Notebook which covers the 3 steps above. Or you can run each step seperately below. To use the Jupyter demo you should additionally install: jupyter, requests, and pytorch-gpu

Training Demo

Download Example Data

First get the multi-scaled training inputs already prepared by running

bash ./scripts/train/get_train_data.sh

Running Training

The example scripts can be found in scripts/train. If using conda env first activate env e.g. conda activate dgts, then from the root directory:

bash ./scripts/train/virus_ball.sh

will train on the spikey-ball from the paper. There is also a demo script for the "sphere rail" and the lizard.

Inference Demo

Get Trained Weights & Some Demo Data

bash ./scripts/inference/get_pretrained_data.sh

Note that if you already ran the training demo from above, this will overwrite some your training snapshots.

Unconditional & Coarse Mesh Generation

This will generate unconditional & conditioned on coarse mesh generative results (Fig. 5 from the paper):

bash ./scripts/inference/anky_generate.sh

Progressively Add Geometric Texture

This will generate a series of progressive textures. The target mesh will progressively gain textures, starting from a low-level, generator, up to a finer resolution generator. This results in a series of animated textures.

bash ./scripts/inference/sphere_rail_animate.sh

Create Training Data Demo

Download Example Data

First get some example 3D meshes with geometric texture

bash ./scripts/gt_optimization/get_demo_data.sh

Running Optimization

The example scripts can be found in scripts/gt_optimization. For example, run the following from the root directory

bash ./scripts/gt_optimization/covid.sh

which will generate the coronavirus from the paper.

Citation

If you find this code useful, please consider citing our paper

@article{Hertz2020deep,
  title = {Deep Geometric Texture Synthesis},
  author = {Hertz, Amir and Hanocka, Rana and Giryes, Raja and Cohen-Or, Daniel},
  year = {2020},
  issue_date = {July 2020}, 
  publisher = {Association for Computing Machinery}, 
  volume = {39}, 
  number = {4}, 
  issn = {0730-0301},
  url = {https://doi.org/10.1145/3386569.3392471},
  doi = {10.1145/3386569.3392471},
  articleno = {108},
  journal = {ACM Trans. Graph.} 
}

Questions / Issues

If you have questions or issues running this code, please open an issue.