Keypoint-Driven Line Drawing Vectorization via PolyVector Flow -- dataset generation

April 9, 2022 · View on GitHub

Example

TLDR: dataset was generated from vector drawings ("Quick, Draw" and "Creative Birds"

Subset: (500 MB, 40 000 images) google drive link Full dataset: (14 GB, 927 000 images) onedrive link

Processing ndjson

See generate_quickdraw.py and generate_quickdraw_svg.py. Given path to ndjson, it dumps corresponding svg file and npz keypoints data to the output folder.

Requirements

Python: ndjson, svgpathtools, shapely, simplification

Data Generation

TLDR: Adobe Illustrator + Adobe ExtendScript
Requirements: Windows, Adobe Illustrator

Download ExtendScript from github https://github.com/Adobe-CEP/CEP-Resources/tree/master/ExtendScript-Toolkit
Javascript documentation for adobe illustrator https://ai-scripting.docsforadobe.dev/

Bugs and Warnings

32-bit only
Data leak detected, rendering speed decreases. We recommend reload the program after 500 images (hence loop limits in the code)
Bug: your template file (template_mybrush.ai) should be cleared

Instructions

Open ExtendScript. Update basePath variable
Link with Illustrator (it will load the program)
Once you run the script, Illustrator may freeze
Observe the progress in Console (top right)

Citation

@article{Puhachov2021KeypointPolyvector,
    author = {Ivan Puhachov and William Neveu and Edward Chien and Mikhail Bessmeltsev},
    title = {Keypoint-Driven Line Drawing Vectorization via PolyVector Flow},
    journal = {ACM Transactions on Graphics (Proceedings of SIGGRAPH Asia)},
    volume = {40}, number = {6}, year = {2021}, month = dec,
    doi = {10.1145/3478513.3480529}
}