Although image generation by deep learning has been developing in recent years, it faces two major challenges. The first is the high cost of training, and the second is the duplication of training data.
Therefore, in this study, we explored a new approach to image generation that is different from the current mainstream methods. The proposed method is to design sketch generation as a reinforcement learning task and solve it with reservoir computing, which is lightweight in terms of training and computation cost.
Although the result was not a perfect sketch generation, it showed new possibilities.