This time, I used carefully selected images to train the model, so it turned out better than last time.