Pictogram is a graphic symbol that uses the figure of an object to express a concept of meaning, and it can express meaning visually without language [1]. So far, still images have been used for pictograms of emergency exit signs and information displays of public facilities, etc. Recently, pictograms augmented by motion are being used, as in the case of the "moving pictograms" [2] that appeared at the Tokyo Olympics 2020. In this study, these pictograms are defined as "motion-pictograms.

Pictograms must be simple and designed for users to understand quickly. However, still-pictograms have limitations. It is necessary to overcome the complexity of the information content and the differences in expression due to nationality and culture. Not everyone can understand the exact meaning at a glance. On the other hand, motion-pictograms can make pictograms more intuitive and accurate understanding by complementing the information.

There are many pictograms in the public and private sectors, including 104 pictograms in the JIS standard pictograms (JIS Z8210) [3], and new pictograms are being created constantly. Compared to still-pictograms, to generate motion-pictograms is more time-consuming, and requires expertise, time and human resources. These costs are one of the reasons why motion-pictograms are not widely used now. Automatically generating of motion-pictograms from still-pictograms make using them more practical and generally way of conveying information.  It is possible to overcome the problems of still-pictograms, as a communication method that does not require language, and to provide a stable and diverse range of information.

In our previous research, the system that recognize figures in textbooks, and generate new 3D object based on the feature value is developed [4]. In this research, it is applied for generating motion-pictogram based on still-pictogram. The developing system uses an AI framework “Liquid Warping GAN with Attention” [5].