Our first community datathon took place this weekend. The dataset was collected by Ilgiz Zhumaev. Ilgiz collected the dataset and processed it for two months; the data is hand-painted letters of the Kyrgyz alphabet. The letters were written in school No. 73 in Bishkek.
Example picture
The main goal of the datathon is to write a model that most accurately recognizes handwritten letters in six hours. Participants were provided with two processed datasets: train (80213 pictures) and validation (8904).
The winner scored 99.09% accuracy on the Accuracy metric using the CNN solution. The top three teams on the leaderboard were awarded.