# Dataset
The dataset contains 4,888 image files (PNG), each associated with a JSON file containing rich annotations including piece bounding boxes and corner coordinates.
Below is a sample image from the dataset.
![sample image][1]
### Extracting
The dataset is split into train (90%), validation (3%), and test (7%) sets.
The training set's ZIP file is over the 5GB limit, so it is split into two files: `train.zip` and `train.z01`. To merge them back into one file, run the following command, after downloading both files: `zip -s 0 train.zip --out train_full.zip`.
Finally, the `train_full.zip`, `val.zip`, and `test.zip` files may be extracted like normal ZIP archives, i.e. `unzip train_full.zip`.
### JSON labels
Each JSON file contains the following information ([this image][2] visualises the labelled information):
- the pixel coordinates of the chessboard's four corner points on the rendered image;
- the FEN description of the pieces on the board;
- the colour of the current player (white/black);
- pixel coordinates of each piece’s bounding box on the rendered image;
- the position and angle of the camera (in the 3D world coordinate system); and
- the lighting mode as well as position, angle, and focus point of each light source in the scene.
The image above is associated with the following JSON file:
{
"fen": "1r3rk1/p3ppbp/6p1/q1p2b2/2P5/4BBP1/P2QPP1P/2R2RK1",
"white_turn": false,
"camera": {
"angle": 45,
"location": [-0.0057182093150913715, 0.2800142765045166, 0.2800142765045166]
},
"lighting": {
"mode": "spotlights",
"flash": { "active": false },
"spot1": {
"active": true,
"xy_angle": 70,
"focus": [-0.23101532459259033, -0.22548413276672363, 0.0],
"location": [0.12312725186347961, 0.338289350271225, 0.21798273921012878]
},
"spot2": {
"active": true,
"xy_angle": 320,
"focus": [-0.05404374375939369, 0.3644087016582489, 0.0],
"location": [0.2757759988307953, -0.23140354454517365, 0.2234768122434616]
}
},
"corners": [
[834, 228],
[978, 665],
[212, 650],
[377, 223]
],
"pieces": [
{ "piece": "k", "square": "g8", "box": [314, 504, 91, 132] },
{ "piece": "r", "square": "f8", "box": [420, 532, 68, 103] },
{ "piece": "r", "square": "b8", "box": [793, 544, 78, 104] },
{ "piece": "p", "square": "h7", "box": [279, 482, 47, 61] },
{ "piece": "b", "square": "g7", "box": [356, 436, 66, 108] },
{ "piece": "p", "square": "f7", "box": [435, 505, 43, 61] },
{ "piece": "p", "square": "e7", "box": [526, 493, 41, 61] },
{ "piece": "p", "square": "a7", "box": [868, 497, 47, 61] },
{ "piece": "p", "square": "g6", "box": [377, 426, 40, 59] },
{ "piece": "b", "square": "f5", "box": [460, 332, 50, 104] },
{ "piece": "p", "square": "c5", "box": [683, 368, 36, 56] },
{ "piece": "q", "square": "a5", "box": [823, 320, 70, 117] },
{ "piece": "P", "square": "c4", "box": [692, 326, 34, 55] },
{ "piece": "P", "square": "g3", "box": [425, 276, 33, 53] },
{ "piece": "B", "square": "f3", "box": [485, 229, 43, 97] },
{ "piece": "B", "square": "e3", "box": [562, 227, 42, 97] },
{ "piece": "P", "square": "h2", "box": [375, 229, 31, 50] },
{ "piece": "P", "square": "f2", "box": [507, 229, 30, 50] },
{ "piece": "P", "square": "e2", "box": [562, 235, 30, 51] },
{ "piece": "Q", "square": "d2", "box": [615, 179, 43, 107] },
{ "piece": "P", "square": "a2", "box": [810, 244, 32, 51] },
{ "piece": "K", "square": "g1", "box": [428, 144, 49, 110] },
{ "piece": "R", "square": "f1", "box": [504, 174, 38, 77] },
{ "piece": "R", "square": "c1", "box": [671, 172, 39, 76] }
]
}
[1]: https://github.com/georgw777/chesscog-report/raw/master/images/3828.png
[2]: https://github.com/georgw777/chesscog-report/blob/master/images/3828_data_synthesis.png
