Method for synthesizing image

A method for synthesizing image comprising:

• a step of preparing a first image and a second image different from the first image, and inputting temporal information and spatial information for each image;
• and
• a step of calculating a recurrent loss (LT) calculated by the following general formula 1 and a recycle loss (Lr) calculated by the following general formula 2 based on the input information, and learning by combining the calculated recurrent loss and recycle loss with an adversarial loss:
• LT(Px)=ΣT−1∥2xt−PX(xt+1)−PX(xt−1)∥2 [General Formula 1]
• Lr(GX,GY,PY)=ΣT−1∥2xt−GX(PY(GY(xt+1)))−GX(PY(GY(xt−1)))∥2 [General Formula 2]
• wherein, xt−1 is a previous time point frame of a specific t time point frame of the first image, xt+1 is a later time point frame of a specific t time point frame of the first image, xt is a specific t time point frame of the first image, P is a prediction function trained to predict a future model from past frames of an image, G is an image generation function, X is the data of the first image, and Y is the data of the second image.