A method for converting an image of a GO game record to training data
Hyeri Yang, Hyeonbeom Heo, Sunguk Jung, Yongwoo Lee, Junyong Hong, Ye Ju Kim, Kyungjae Lee · 2022
Supervised learning is an approach to training artificial intelligence using a labeled dataset. Since the interest in artificial intelligence is high, preparing training data for supervised learning has an important meaning. In this paper, we propose a feasible algorithm for converting a GO game dataset stored as an image into learnable labeled data. The training data need to have the location and color of GO stones. However, detecting a GO board is difficult because of the area covered by a GO stone. Furthermore, it is hard to know the reference point of the board, so recognizing the locations of the stone is also difficult. To detect a GO board, the proposed method is applied differently depending on whether or not there are GO stones on the board. Additionally, by finding the starting point of the GO board, we can convert the location of the GO stone in the image to training data.