Skip to content

SketchyScene/SketchyCOCO

 
 

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

15 Commits
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

SketchyCOCO dataset

This repository hosts the ShetchyCOCO dataset. Please refer to our CVPR paper for more information: "SketchyCOCO: Image Generation from Freehand Scene Sketches".

SketchyCOCO dataset can be explored by https://sysu-imsl.com/EdgeGAN/dataset.html.

Overview

Dataset

SketchyCOCO dataset consists of two part:

  • Object-level data
    • Object-level data contains $20198(train18869+val1329)$ triplets of {foreground sketch, foreground image, foreground edge map} examples covering 14 classes, $27683(train22171+val5512)$ pairs of {background sketch, background image} examples covering 3 classes.
  • Scene-level data
    • Scene-level data contains $14081(train 11265 + val 2816)$ pairs of {foreground image&background sketch, scene image} examples, $14081(train 11265 + val 2816)$ pairs of {scene sketch, scene image} examples and the segmentation ground truth for $14081(train 11265 + val 2816)$ scene sketches. Some val scene images come from the train images of the COCO-Stuff dataset for increasing the number of the val images of the SketchyCOCO dataset.

Dataset Augmentaion

We increase 4662 images for 9 objects, their correspondence edge maps and sketches. The details of dataset augmentation is shown below. And the augmentation has been merged into Object-level data.

cat dog zebra giraffe horse cow elephant sheep Car
659 777 401 246 773 628 398 369 411

Captions of the file structure in the dataset

  • data
    • Scene - Scene-level data
      • GT - Ground Truth
        • trainInTrain - Train images of SketchyCOCO dataset from the train images of the COCO-Stuff dataset
        • valInTrain - Val images of SketchyCOCO dataset from the train images of the COCO-Stuff dataset
        • val - Val images of SketchyCOCO dataset from the val images of the COCO-Stuff dataset
      • Sketch - Sketch scene of GT (a sketch scene has the same name with the corresponding GT)
      • Annotation - Annotations for sketch scene segmentation
    • Object - Object-level data
      • GT - Ground Truth
      • Sketch - Sketch image of the GT (a edge image has the same name with the corresponding GT)
      • Edge - Edge image of the GT (a edge image has the same name with the corresponding GT)
    • Others - Intermediate products when building the dataset
      • background - Background images and sketches
      • background_training - Images of {foreground image&background sketch} data
      • foreground - Foreground images and sketches used in the scene
      • intermediate product - Images of {generated image&background sketch} data
      • sketches_background - Sketches for building the background sketches
    • Image Source - Files storing the source of images
  • matlab_code - Codes for building the dataset

Google Drive Hosting(in preparing)

Optional

  • You can build a new dataset using the following instructions:

    1. Install COCO API for Matlab.

    2. Download images and annotations of the COCO-Stuff dataset.

    3. Run ./matlab_code/constructDataset.m after changing the parameters in the code.

  • The XDoG image can be obtained by running ./matlab_code/XDoG.m after changing the parameters in the code.

  • The pairs of the objects can be obtained by running ./matlab_code/preprocess_combine.m after changing the parameters in the code.

  • The pairs of the scenes can be obtained by running ./matlab_code/combineScript.m after changing the parameters in the code.

Licensing

SketchyCOCO is a derivative work of [the Sketchy Database](http://sketchy.eye.gatech.edu/), [the COCO-Stuff Dataset](https://github.com/nightrome/cocostuff), [TU Berlin](http://cybertron.cg.tu-berlin.de/eitz/projects/classifysketch/), [Car Dataset](http://ai.stanford.edu/~jkrause/cars/car_dataset.html), [cats_vs_dogs](https://tensorflow.google.cn/datasets/catalog/cats_vs_dogs?hl=zh-cn) and [Pascal voc 2012](http://host.robots.ox.ac.uk/pascal/VOC/). The authors of the COCO-Stuff database, the Sketchy Database, TU Berlin, Car Dataset, cats_vs_dogs and Pascal voc 2012 do not in any form endorse this work. Different licenses apply:

About

SketchyCOCO dataset (CVPR 2020)

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • MATLAB 79.7%
  • HTML 10.5%
  • JavaScript 5.4%
  • CSS 4.4%