US 12,175,573 B2
Automatic design-creating artificial neural network device and method, using UX-bits
Joon Yun Choi, Gwangju-si (KR)
Assigned to TINTOLAB CO., LTD, Seoul (KR)
Appl. No. 18/017,349
Filed by TINTOLAB CO., LTD, Seoul (KR)
PCT Filed Apr. 15, 2022, PCT No. PCT/KR2022/005464
§ 371(c)(1), (2) Date Jan. 20, 2023,
PCT Pub. No. WO2022/255632, PCT Pub. Date Dec. 8, 2022.
Claims priority of application No. 10-2021-0070403 (KR), filed on May 31, 2021.
Prior Publication US 2023/0351655 A1, Nov. 2, 2023
Int. Cl. G06T 11/60 (2006.01); G06F 9/451 (2018.01); G06T 3/4046 (2024.01)
CPC G06T 11/60 (2013.01) [G06F 9/451 (2018.02); G06T 3/4046 (2013.01)] 8 Claims
OG exemplary drawing
 
1. An automatic design generating artificial neural network device using a user experience (UX)-bit, comprising:
an image theme encoding module that is an encoding module which receives image theme data, which is an image representing a theme of a web/app graphic design to be generated by a practitioner, as input data, and outputs an image theme encoding vector as output data;
a text theme encoding module that is an encoding module which receives text theme data, which is text representing the theme of the web/app graphic design to be generated by the practitioner, as input data, and outputs a text theme encoding vector as output data;
a UX-bit generation module that is a module which receives the image theme encoding vector and the text theme encoding vector as input data and outputs UX-bit attributes of a plurality of UX elements as output data;
a design generation module that is an upsampling artificial neural network module which receives the image theme encoding vector, the text theme encoding vector, and the UX-bit attribute as input data and outputs design data indicating a specific web/app graphic design as output data;
an image theme discriminator that is a module used in a learning session of the design generation module and is a pre-learned artificial neural network module which, when the design data and the image theme data are input as input data, outputs an image theme discrimination vector, which indicates a probability of similarity between the design data and the image theme data, as output data;
a text theme discriminator that is a module used in the learning session of the design generation module and is a pre-learned artificial neural network module which, when a design encoding vector that is an encoding vector of the design data and the text theme encoding vector are input as input data, outputs a text theme discrimination vector, which indicates a probability of similarity between the design encoding vector and the text theme encoding vector, as output data; and
a UX-bit attribute discriminator that is a module used in the learning session of the design generation module and is a pre-learned artificial neural network module which, when the design encoding vector and an encoding vector of the UX-bit attribute are input as input data, outputs a UX-bit attribute discrimination vector, which indicates a probability of similarity between the design encoding vector and the encoding vector of the UX-bit attribute, as output data,
wherein, in the learning session of the design generation module, parameters of the design generation module are updated in a direction in which a representative design loss, which is composed of a difference between the design data and web/app design reference data (ground truth) in which similarity with the image theme encoding vector and similarity with the text theme encoding vector are greater than or equal to a specific level in an encoding vector of a pre-stored web/app design corresponding thereto, an image theme discrimination loss comprising the image theme discrimination vector, a text theme discrimination loss comprising the text theme discrimination vector, and a UX-bit attribute determination loss comprising the UX-bit attribute discrimination are reduced.