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 or https://sysu-imsl.github.io/EdgeGAN/dataset.html.
- Dataset
- Dataset Augmentation
- Captions of the file structure in the dataset
- Google Drive Hosting
- Baidu Netdisk Hosting
- Optional
- Licensing
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.
- Object-level data contains
- 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.
- Scene-level data contains
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 |
- 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
- GT - Ground Truth
- 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
- Scene - Scene-level data
- matlab_code - Codes for building the dataset
- Object-level data Password:nv6n
- Scene-level data A Password:7k48
- Scene-level data B Password:l43w
- Scene-level data C Password:3umq
- Others Password:4hy0
- Image Source Password:vrem
PS: Merge trainInTrain_part of Scene-level data B and data C into GT/trainInTrain of Scene-level data A after downloading.
-
You can build a new dataset using the following instructions:
-
Install COCO API for Matlab.
-
Download images and annotations of the COCO-Stuff dataset.
-
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.
- Images: Flicker Terms of use
- Sketches: Creative Commons Attribution 4.0 License
- SketchyCOCO code: Creative Commons Attribution 4.0 License