Football360

This is the official repository for the Football360 dataset. The aim of this dataset is to aid the development and evaluation of computer vision algorithms, primarily the radial distortion correction algorithms, in the sports domain. The work presented here was published in the VISAPP 2023 conference.

The original paper can be found at - https://www.scitepress.org/PublishedPapers/2023/116812/116812.pdf.

This dataset contains 268 panorama images, and was created using the PANONO panoramic camera in 3 football arenas in Slovakia. Each arena was covered from numerous locations on all levels of the tribune, and broadcast camera platforms. The images capture regular football game, pitch maintenance, low/challenging lighting conditions, day and night situations.

Arena	Number of images
Bratislava	111
Trnava	104
Dunajska Streda	53

Download the raw images from the following link:

File	Images	Download Link
Raw Data	268	football360-raw.tar.gz (7.7 GB)

Raw Image Probe

Raw images are stored as 16384x8192 JPG, they are the direct result of the PANONO stitching service.

Pinhole Image Probe

From the panorama images a high number of crop images conforming to the pinhole camera model can be generated by randomly sampling the camera orientation and focal length. If desired, the generated images can be distorted with a chosen radial distortion model to produce very large training set suitable for neural network training.

Left	Front	Right

Distortion Probe

The exporting process supports both polynomial distortion model with two parameters. The probability distributions of the distortion parameters can be configured by a preset JSON file.

Barrel	Pincushion

Exported Sets

The exported datasets with their respective configurations can be found in the following table.

Set name	Purpose	Images	Preset	Download Link
A	Training	30,000	setA.json	football360-setA.h5 (10.5 GB)
B	Training	100,000	setB.json	football360-setB.h5 (35.2 GB)
C	Training	300,000	setC.json	football360-setC.h5 (105.5 GB)
V	Validation	10,000	setV.json	football360-setV.h5 (3.5 GB)

Exporting Process

The exporting process consists of two steps:

• Splitting of the input image folder
• Rendering cropped images according to the split subsets

To split the input image folder execute the following command:

./exporter -in IMAGES_FOLDER -split 90;10 -os split.json

To render the cropped distorted images using a preset execute the following command. Make sure to select either training or validation subset for output.

./exporter -in IMAGES_FOLDER -is split.json -ip presets/setA.json \
           -ot TRAINING_IMAGES.H5

./exporter -in IMAGES_FOLDER -is split.json -ip presets/setA.json \
           -ov VALIDATION_IMAGES.H5

Presets

Presets are defined as JSON files, and they contain the parameters relevant to the export process

{
    "renderSize": [1920, 1080],
    "scaleSize": [448, 448],
    "compression": "png",
    "nImages": 30000,    
    "view": {
        "pan": [-40, 40],
        "tilt": [-25, -2],
        "roll": [-2, 2],
        "fov": [10, 50]
    },
    "distortion": "poly-2p",
    "distortionParams": {
        "k1": [-0.45, 0.12],
        "epsK2": 0.02
    }
}

Citing Football360

If you find Football360 useful in your research, please consider citing:

@conference{janos2023football,
    author={Igor Jánoš. and Vanda Benešová.},
    title={Football360: Introducing a New Dataset for Camera Calibration in Sports Domain},
    booktitle={Proceedings of the 18th International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications (VISIGRAPP 2023) - Volume 4: VISAPP},
    year={2023},
    pages={301-308},
    publisher={SciTePress},
    organization={INSTICC},
    doi={10.5220/0011681200003417},
    isbn={978-989-758-634-7},
    issn={2184-4321},
}