Skip to content

Latest commit

 

History

History
66 lines (59 loc) · 2.45 KB

README.md

File metadata and controls

66 lines (59 loc) · 2.45 KB
Raw

Neural Strokes: Stylized Line Drawing of 3D Shapes

This repository contains the PyTorch implementation for ICCV 2021 Paper "Neural Strokes: Stylized Line Drawing of 3D Shapes" by Difan Liu, Matthew Fisher, Aaron Hertzmann, Evangelos Kalogerakis.

Dependency

  • The project is developed on Ubuntu 16.04 with cuda10.0 + cudnn7.6. The code has been tested with PyTorch 1.2.0 (GPU version) and Python 3.7.6.
  • Python packages:
    • diffvg
    • OpenCV (tested with 4.3.0)
    • PyYAML (tested with 5.3.1)
    • scikit-image (tested with 0.17.2)
    • scipy (tested with 1.4.1)

Dataset and Weights

  • Pre-trained model is available here, please put it in weights:

    cd weights
unzip weights.zip

  • Dataset is available here, please put it in datasets:

    cd datasets
unzip datasets.zip

Also refer to https://github.com/Chanuku/semi_ref2sketch_code

Preprocessing

  • Preprocess raw data: 
    python preprocess.py -s datasets/style_01
    -s: path to the data of a single style.

Test

  • Test: 
    python test.py -d datasets/style_01/test_2 -g weights/style_01/SG.pth -t weights/style_01/ST.pth -s results/style_01__test_2.png
    -d: path to testing data.
    -g: path to Stroke Geometry checkpoint.
    -t: path to Stroke Texture checkpoint.
    -s: path to save the synthesized image.

Training

  • Start the Stroke Geometry (SG) training: 
    python train_SG.py -d datasets/style_01/train -n style_01__SG
  • After the Stroke Geometry (SG) training is finished, start the Stroke Texture (ST) training: 
    python train_ST.py -d datasets/style_01/train -n style_01__ST
    -d: path to training data.
    -n: name of the experiment.

Cite:

@InProceedings{Liu_2021_ICCV,
author={Liu, Difan and Fisher, Matthew and Hertzmann, Aaron and Kalogerakis, Evangelos},
title={Neural Strokes: Stylized Line Drawing of 3D Shapes},
booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision},
year = {2021}
}

Contact

To ask questions, please email.