-
Notifications
You must be signed in to change notification settings - Fork 42
en_IkaLogChallenge
Today 4200+ users signed up on stat.ink, and terabytes of Splatoon scoreboard images are processed with IkaLog. The images include numbers and text that possible to get OCRed, and icon images can be classified. IkaLog Challenge provides the dataset for evaluation and research purpose that provided from the another repository.
IkaLog Challenge dataset includes the image data generated by Nintendo's Splatoon game, but it never includes the original software and its data from the copy.
Weapon Classification dataset for weapon icon image classification that IkaLog tries to solve. It includes 150,000+ weapon icon images that can be classified to couples of dozens of classes - more than 80. The images are pre-classified to sub directories with (maybe) 99.9% of accuracy.
The images are captured from HDMI signal from the game console, gathered without compromising Nintendo's IP. The each image can be always different because the game renders in different form every time (depends on the inklings' headwear and team color), HDMI capture devices and its device drivers apply their specific imagee processing. In addition, some IkaLog environments are not configured as expected. As the result we have many weapon icon images with noise. The image is somewhat smaller than what typical image recognition techniques expects.
To get the data set, Visit here