A code repository associated with the paper A Benchmark for Rough Sketch Cleanup by Chuan Yan, David Vanderhaeghe, and Yotam Gingold from SIGGRAPH Asia 2020.

Overview

A Benchmark for Rough Sketch Cleanup

This is the code repository associated with the paper A Benchmark for Rough Sketch Cleanup by Chuan Yan, David Vanderhaeghe, and Yotam Gingold from SIGGRAPH Asia 2020.

This code computes the metrics described in the paper and generates the benchmark website to compare the output of various sketch cleanup algorithms.

The Directory Structure

Data directories are defined in the file cfg.yaml:

  • dataset_dir: User puts the dataset here. Needed by the website.
  • alg_dir: User puts automatic results here. Needed by the website.
  • web_dir: We generate the website here. Image paths look like ../{alg_dir}/rest/of/path.svg
  • table_dir: We generate the metrics computed by the benchmark here. Needed to generate the website, but not needed when hosting the website. (A precomputed version for algorithms we tested is provided below.)
  • test_dir: We generate resized image files for testing algorithms here. Needed also when computing metrics. Not needed by the website. (A precomputed version is provided below.)

The default values are

dataset_dir: './data/Benchmark_Dataset
alg_dir: './data/Automatic_Results'
web_dir: './data/web'
table_dir: './data/Evaluation_Data'
test_dir: './data/Benchmark_Testset'

If you are generating your own test_dir data, you need Inkscape and ImageMagick. run_benchmark.py tries to find them according to your OS. You can set the paths directly in cfg.yaml by changing inkscape_path and magick_path to point to Inkscape and ImageMagick's convert executable, respectively.

Installing Code Dependencies

Clone or download this repository. The code is written in Python. It depends on the following modules: aabbtree, CairoSVG, cssutils, matplotlib, numpy, opencv-python, pandas, Pillow, PyYAML, scipy, svglib, svgpathtools, tqdm

You can install these modules with:

pip3 install -r requirements.txt

or, for a more reproducible environment, use Poetry (brew install poetry or pip install poetry):

poetry install --no-root
poetry shell

or Pipenv (pip install pipenv):

pipenv install
pipenv shell

The shell command turns on the virtual environment. It should be run once before running the scripts.

If you are not downloading the precomputed test images, make sure the following external software has been installed in your system:

  1. Inkscape 1.x. Please install an up-to-date Inkscape. Versions prior to 1.0 have incompatible command line parameters. brew cask install inkscape or apt-get install inkscape.
  2. ImageMagick. brew install imagemagick or apt-get install imagemagick.

The Dataset and Precomputed Output

You can download the sketch dataset, precomputed algorithmic output, and computed metrics here: Benchmark_Dataset.zip (900 MB), Automatic_Results.zip (440 MB), Evaluation_Data.zip (20 MB). Unzip them in ./data/ (unless you changed the paths in cfg.yaml):

unzip Benchmark_Dataset.zip
unzip Automatic_Results.zip
unzip Evaluation_Data.zip

Note that the vectorized data has been normalized to have uniform line width. It was too tedious for artists to match line widths with the underlying image, so we did not require them to do so and then normalized the data.

Running

Generating or Downloading the Testset

(If you are trying to regenerate the website from the paper using the precomputed output and already computed metrics, you do not need the Testset. If you want to change anything except the website itself, you need it.)

The Testset consists of files derived from the dataset: rasterized versions of vector images and downsized images. You can regenerate it (see below) or download Benchmark_Testset.zip (780 MB) and extract it into ./data/ (unless you changed the paths in cfg.yaml):

unzip Benchmark_Testset.zip

You can regenerate the Testset (necessary if you change the dataset itself) by running the following commands:

python3 run_benchmark.py --normalize   # generate normalized versions of SVGs
python3 run_benchmark.py --generate-test # generate rasterized versions of Dataset, at different resolutions

This will scan dataset_dir and test_dir, generate missing normalized and rasterized images as needed. It takes approximately 20 to 30 minutes to generate the entire Testset.

Adding Algorithms to the Benchmark

Run your algorithm on all images in the Testset. If your algorithm takes raster input, run on all images in ./data/Benchmark_Testset/rough/pixel. If your algorithm takes vector input, run on all images in ./data/Benchmark_Testset/rough/vector. For each input, save the corresponding output image as a file with the same name in the directory: ./data/Automatic_Results/{name_of_your_method}{input_type}/{parameter}/

The algorithm folder name must contain two parts: name_of_your_method with an input_type suffix. The input_type suffix must be either -png or -svg. The parameter subdirectory can be any string; the string none is replaced with the empty string when generating the website. Folders beginning with a . are ignored. For examples, see the precomputed algorithmic output in ./Automatic_Results. and evaluation result in ./Evaluation_Data already.

If your algorithm runs via alg path/to/input.svg path/to/output.png, here are two example commands to run your algorithm in batch on the entire benchmark. Via find and parallel

find ./data/Benchmark_Testset/rough/pixel -name '*.png' -print0 | parallel -0 alg '{}' './data/Automatic_Results/MyAlgorithm-png/none/{/.}.svg'

Via fd:

fd ./data/Benchmark_Testset/rough/pixel -e png -x alg '{}' './data/Automatic_Results/MyAlgorithm-png/none/{/.}.svg'

Computing the Metrics

Run the evaluation with the command:

python3 run_benchmark.py --evaluation

This command creates CSV files in ./data/Evaluation_Data. It will not overwrite existing CSV files. If you downloaded the precomputed data, remove a file to regenerate it.

Generating the Website to View Evaluation Results

After you have called the evaluation step above to compute the metrics, generate the website with the command:

python3 run_benchmark.py --website

You must also generate thumbnails once with the command:

python3 run_benchmark.py --thumbs

Internally, the --thumbs command creates a shell that calls find, convert, and parallel.

To view the website, open the help.html or index.html inside the web_dir manually or else call:

python3 run_benchmark.py --show

The website visualizes all algorithms' output and plots the metrics.

Putting It All Together

If you don't want to call each step separately, simply call:

python3 run_benchmark.py --all

Computing Metrics on a Single Sketch

Similarity Metrics

To run the similarity metrics manually, use tools/metric_multiple.py. To get help, run:

python3 tools/metric_multiple.py --help

To compare two files:

python3 tools/metric_multiple.py -gt "example/simple-single-dot.png" -i "example/simple-single-dot-horizontal1.png" -d 0 --f-measure --chamfer --hausdorff

Vector Metrics

To evaluate junction quality:

python3 tools/junction_quality.py --help

To compute arc length statistics:

python3 tools/svg_arclengths_statistics.py --help

Rasterization

If you need to convert a file from an SVG to a PNG, you can do it specifying the output filename:

inkscape my_file.svg --export-filename="output-WIDTH.png" --export-width=WIDTH --export-height=HEIGHT

or specifying the output type (the input filename's extension is replaced):

inkscape my_file.svg --export-type=png --export-width=WIDTH --export-height=HEIGHT

The shorthand versions of the above rasterization commands are:

inkscape -o output-WIDTH.png -w WIDTH -h HEIGHT my_file.svg

or

inkscape --export-type=png -w WIDTH -h HEIGHT my_file.svg

If you pass only one of width or height, the other is chosen automatically in a manner preserving the aspect ratio.

Companion code for the paper "Meta-Learning the Search Distribution of Black-Box Random Search Based Adversarial Attacks" by Yatsura et al.

META-RS This is the companion code for the paper "Meta-Learning the Search Distribution of Black-Box Random Search Based Adversarial Attacks" by Yatsu

Bosch Research 7 Dec 09, 2022
Pairwise Learning for Neural Link Prediction for OGB (PLNLP-OGB)

Pairwise Learning for Neural Link Prediction for OGB (PLNLP-OGB) This repository provides evaluation codes of PLNLP for OGB link property prediction t

Zhitao WANG 31 Oct 10, 2022
This repo is customed for VisDrone.

Object Detection for VisDrone(无人机航拍图像目标检测) My environment 1、Windows10 (Linux available) 2、tensorflow = 1.12.0 3、python3.6 (anaconda) 4、cv2 5、ensemble

53 Jul 17, 2022
A custom-designed Spider Robot trained to walk using Deep RL in a PyBullet Simulation

SpiderBot_DeepRL Title: Implementation of Single and Multi-Agent Deep Reinforcement Learning Algorithms for a Walking Spider Robot Authors(s): Arijit

Arijit Dasgupta 9 Jul 28, 2022
Code and Data for NeurIPS2021 Paper "A Dataset for Answering Time-Sensitive Questions"

Time-Sensitive-QA The repo contains the dataset and code for NeurIPS2021 (dataset track) paper Time-Sensitive Question Answering dataset. The dataset

wenhu chen 35 Nov 14, 2022
An implementation of the [Hierarchical (Sig-Wasserstein) GAN] algorithm for large dimensional Time Series Generation

Hierarchical GAN for large dimensional financial market data Implementation This repository is an implementation of the [Hierarchical (Sig-Wasserstein

11 Nov 29, 2022
Official implementation of the paper Visual Parser: Representing Part-whole Hierarchies with Transformers

Visual Parser (ViP) This is the official implementation of the paper Visual Parser: Representing Part-whole Hierarchies with Transformers. Key Feature

Shuyang Sun 117 Dec 11, 2022
E2e music remastering system - End-to-end Music Remastering System Using Self-supervised and Adversarial Training

End-to-end Music Remastering System This repository includes source code and pre

Junghyun (Tony) Koo 37 Dec 15, 2022
Official implementation for the paper: Permutation Invariant Graph Generation via Score-Based Generative Modeling

Permutation Invariant Graph Generation via Score-Based Generative Modeling This repo contains the official implementation for the paper Permutation In

64 Dec 29, 2022
Use stochastic processes to generate samples and use them to train a fully-connected neural network based on Keras

Use stochastic processes to generate samples and use them to train a fully-connected neural network based on Keras which will then be used to generate residuals

Federico Lopez 2 Jan 14, 2022
Parameter-ensemble-differential-evolution - Shows how to do parameter ensembling using differential evolution.

Ensembling parameters with differential evolution This repository shows how to ensemble parameters of two trained neural networks using differential e

Sayak Paul 9 May 04, 2022
An implementation of based on pytorch and mmcv

FisherPruning-Pytorch An implementation of Group Fisher Pruning for Practical Network Compression based on pytorch and mmcv Main Functions Pruning f

Peng Lu 15 Dec 17, 2022
a delightful machine learning tool that allows you to train, test and use models without writing code

igel A delightful machine learning tool that allows you to train/fit, test and use models without writing code Note I'm also working on a GUI desktop

Nidhal Baccouri 3k Jan 05, 2023
Pytorch port of Google Research's LEAF Audio paper

leaf-audio-pytorch Pytorch port of Google Research's LEAF Audio paper published at ICLR 2021. This port is not completely finished, but the Leaf() fro

Dennis Fedorishin 80 Oct 31, 2022
A Simple Example for Imitation Learning with Dataset Aggregation (DAGGER) on Torcs Env

Imitation Learning with Dataset Aggregation (DAGGER) on Torcs Env This repository implements a simple algorithm for imitation learning: DAGGER. In thi

Hao 66 Nov 23, 2022
An example project demonstrating how the Autonomous Learning Library can be used to build new reinforcement learning agents.

About This repository shows how Autonomous Learning Library can be used to build new reinforcement learning agents. In particular, it contains a model

Chris Nota 5 Aug 30, 2022
Negative Interactions for Improved Collaborative Filtering:

Negative Interactions for Improved Collaborative Filtering: Don’t go Deeper, go Higher This notebook provides an implementation in Python 3 of the alg

Harald Steck 21 Mar 05, 2022
Face and Body Tracking for VRM 3D models on the web.

Kalidoface 3D - Face and Full-Body tracking for Vtubing on the web! A sequal to Kalidoface which supports Live2D avatars, Kalidoface 3D is a web app t

Rich 257 Jan 02, 2023
On-device wake word detection powered by deep learning.

Porcupine Made in Vancouver, Canada by Picovoice Porcupine is a highly-accurate and lightweight wake word engine. It enables building always-listening

Picovoice 2.8k Dec 29, 2022
A set of tools for converting a darknet dataset to COCO format working with YOLOX

darknet格式数据→COCO darknet训练数据目录结构(详情参见dataset/darknet): darknet ├── class.names ├── gen_config.data ├── gen_train.txt ├── gen_valid.txt └── images

RapidAI-NG 148 Jan 03, 2023