Details
GVC Puzzles Dataset (GVCPuzzles) is a large collection of synthesized jigsaw puzzles for different
type of images, along with the ground-truth transformations and the tools to manipulate and display them.
The database was funded by National Science Foundation (NSF) CISE IIS-1320959.
GVCPuzzles has large diversities on both the type of images and the difficulty level of the puzzle, including
- 5 categories of images
- animal: 134 samples.
- art: 146 samples
- building: 114 samples.
- landscape: 145 samples.
- portrait: 99 samples.
- 6 difficulty levels
- Level 1: less than 20 pieces ~ 20%.
- Level 2: between 20 to 39 pieces ~ 30%.
- Level 3: between 40 to 59 pieces ~ 30%.
- Level 4: between 60 to 79 pieces ~ 10%.
- Level 5: between 80 to 89 pieces ~ 7%.
- Level 6: larger than 89 pieces ~ 3%.
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
Directory Tree of the original image files:
images
├── animal
│ ├── 0j38qB6Js4I.jpg
│ ├── 0Sd2qqU5soQ.jpg
│ ├── 17FcdBbY00M.jpg
│ ├── 19MqG0yRIDc.jpg
│ └── ...
├── art
│ ├── 0igZnMr6ukk.jpg
│ ├── 0V3qHmSV1kM.jpg
│ ├── 13_ztzChCNU.jpg
│ ├── 1fs5DIRsuEk.jpg
│ └── ...
├── building
│ ├── 0nnuhlN0BSY.jpg
│ ├── 1ak3Z7ZmtQA.jpg
│ ├── 1nlhZy86uPU.jpg
│ ├── 1PHDS-PFtcM.jpg
│ └── ...
├── landscape
│ ├── 0ALhqKA4XTc.jpg
│ ├── 0o_GEzyargo.jpg
│ ├── 0x6RTts1jRU.jpg
│ ├── 13VwsTt9pAw.jpg
│ └── ...
└── portrait
├── 0ZYPu-nLOwU.jpg
├── 11f9F8I6VcA.jpg
├── 13_7xGYBb7c.jpg
├── 14XzQZ6I3nI.jpg
└── ...
Directory Tree of the puzzles:
puzzles
├── animal
│ ├── [L1C14]0j38qB6Js4I
│ │ ├── config.txt
│ │ ├── groundtruth.json
│ │ ├── groundtruth.txt
│ │ ├── piece-0.png
│ │ ├── piece-1.png
│ │ ├── piece-2.png
│ │ ├── piece-3.png
│ │ └── ...
│ ├── [L3C55]0Sd2qqU5soQ
│ │ ├── config.txt
│ │ ├── groundtruth.json
│ │ ├── groundtruth.txt
│ │ ├── piece-0.png
│ │ ├── piece-1.png
│ │ ├── piece-2.png
│ │ ├── piece-3.png
│ │ └── ...
│ └── ...
├── art
│ ├── [L2C27]0igZnMr6ukk
│ │ ├── config.txt
│ │ ├── groundtruth.json
│ │ ├── groundtruth.txt
│ │ ├── piece-0.png
│ │ ├── piece-1.png
│ │ ├── piece-2.png
│ │ ├── piece-3.png
│ │ └── ...
│ └── ...
└── ...
Some details of puzzles directory:
The name of the directory of a puzzle has the format as: [LxCy]puzzle-id.x is the difficulty level of this puzzle. y is the pieces number of this puzzle. puzzle-id is the name of the image which corresponding to puzzle-id.jpg in the image files under the same category. For example, the puzzle [L3C55]0Sd2qqU5soQ in the animal means this puzzle has level 3 difficulty and 55 pieces fragments. The complete image for this puzzle is 0Sd2qqU5soQ.jpg in the original image files under animal category.
For each puzzle, there are some extra files such as config.txt, groundtruth.json and groundtruth.txt. They describe the solution of this puzzle, which is the transform matrix for each fragment that could restore the image. For more specific details of the groundtruth transformation matrix, please refer to File formats.
Puzzle Visualization tool
We also provide a visualization tool which could visualize the fragments in the puzzle. Please refer to PuzzleDebugger.Downloads
-
Example Images -
Example Puzzles
You can download a example set of the dataset through above links (3 samples in each category).
To gain access to the complete dataset, please send your request to xinli@cct.lsu.edu
Note
- The GVCPuzzles dataset is available for non-commercial research purposes only.
- All images of the GVCPuzzles dataset are obtained from the Internet which are not property of GVC, LSU. The GVC is not responsible for the content nor the meaning of these images.
- You agree not to reproduce, duplicate, copy, sell, trade, resell or exploit for any commercial purposes, any portion of the images and any portion of derived data.
- You agree not to further copy, publish or distribute any portion of the GVCPuzzles dataset. Except, for internal use at a single site within the same organization it is allowed to make copies of the dataset.
- The GVC reserves the right to terminate your access to the GVCPuzzles dataset at any time.
Citation
 |
JigsawNet: Shredded Image Reassembly Using Convolutional Neural Network and Loop-based Composition.
C. Le and X. Li. IEEE Transactions on Image Processing, to appear, 2019. [Bibtex] [Paper] |
 |
Hierarchical Fragmented Image Reassembly using a Bundle-of-Superpixel Representation
X. Li, K. Xie, W. Hong, C. Liu Computer-aided Geometric Design (CAGD), International Conf. on Geometric Modeling and Processing (GMP19), to appear, 2019 [Bibtex] [Paper] |