Sketch-Generation-with-Drawing-Process-Guided-by-Vector-Flow-and-Grayscale

This is the official implementation of the AAAI 2021: AAAI Conference on Artificial Intelligence accepted paper "Sketch Generation with Drawing Process Guided by Vector Flow and Grayscale".
Our paper's preprint version is available at arXiv now: https://arxiv.org/abs/2012.09004.
Our Supplementary Material (PDF) is available at Baidu Netdisk (百度网盘) https://pan.baidu.com/s/1ZZEmpuPXRNBPvG0WHbbPKA. The extraction code (提取码) is 1234.
Here we give some instructions for running our code.

Authors

First Author

Zhengyan Tong (童峥岩)， 此论文idea提供者、代码原作者、论文主笔者，主要从事计算机视觉方面的研究。发表此论文时为上海交通大学电子信息与电气工程学院信息工程专业大四在读本科生。联系方式: [email protected]

Other Authors

• Xuanhong Chen (Shanghai Jiao Tong University Ph. D.)
• Bingbing Ni (Shanghai Jiao Tong University Associate Professor) Corresponding author
• Xiaohang Wang (Shanghai Jiao Tong University Undergraduate)

Acknowledgments

• I am extremely grateful to the Second Author Xuanhong Chen for his professional advice, comments, and encouragement, which greatly improves this work.
• In particular，I would like to express my gratitude to my junior high school classmate Tianhao Shen for his enthusiastic and selfless help in code debugging, although he is not one of the authors of this paper.

Examples

We give three examples that can be run directly (the hyperparameters of these three examples have been fixed).

Quick start

• To draw the cat: python cat.py
• To draw the dog: python dog.py
• To draw the girl: python girl.py

Results

Instructions

To draw arbitrary input: python process_order.py. Of course you need to adjust the following parameters.

Hyperparameters

• input_path = './input/your file' Input image path
• output_path = './output' Do not change this
• n = 10 Gray-scale quantization order
• period = 5 Line(stroke) width
• direction = 10 Direction quantization order
• Freq = 100 Save the drawing process every Freq strokes are drawn
• deepen = 1 Edge map's intensity. The bigger, the darker.
• transTone = False Do not change this
• kernel_radius = 3 Edge tangent flow kernel size, do not change this
• iter_time = 15 Edge tangent flow kernel iterations times, do not change this
• background_dir = None Whether fix the drawing direction in the background, this value could be None or an integer between (0~180)
• CLAHE = True Whether input uses CLAHE (Do not change this)
• edge_CLAHE = True Whether edge map uses CLAHE (Do not change this)
• draw_new = True Do not change this
• random_order = False Use random drawing order if True
• ETF_order = True Use the drawing order described in our paper if True
• process_visible = True Whether show the drawing process

In our supplementary material (PDF), we explain these hyperparameters in more detail and we show more comparisons with existing pencil drawing algorithms. We also offer more results of our method. Our Supplementary Material is available at Baidu Netdisk (百度网盘) https://pan.baidu.com/s/1ZZEmpuPXRNBPvG0WHbbPKA. The extraction code (提取码) is 1234.

Results