Technology Inheritance Meaʻai Autū: Meafaigaluega Fatuina o Mamanu Fausiaga e Junming Chen

O le mamanu Technology Inheritance Meaʻai Autū: Meafaigaluega Fatuina o Mamanu Fausiaga e Junming Chen o se manumalo maualuga tele o le fa'ailoga mamalu 2023 A' Design Award Uʻamea i le Vaega Tauva Mamanu Fa'atupu.

UʻameaMamanu Fa'atupu· 2023

Technology Inheritance Meaʻai Autū: Meafaigaluega Fatuina o Mamanu Fausiaga

na mamanuina e Junming Chen

Tagata Fa'atau: Chen junming

Manumalo o le A' Design Award Uʻamea i le Vaega Mamanu Fa'atupu

45,702

Manumalo Fa'ailoga Fa'amaonia

Fa'amaonia le Tulaga o le Fa'ailoga mo #150834

Va'ai i le A' Design Award Fa'asalalauga Pailesi o le Fa'ailoga

Lanu

Le Mamanu

E Uiga ia Technology Inheritance Meaʻai Autū: Meafaigaluega Fatuina o Mamanu Fausiaga

Tusitusiga o le Tusi: O le Talatuu Tekonolosi o se meafaigaluega atamai (artificial intelligence) lea e faia ai mamanu i fafo o fale. O le meafaigaluega e fatuina ai foliga o fale i ituaiga eseese o le fausiaga mai upu ua tuuina atu e le tagata faaaoga. E fatuina lava le ata e tasi i ni nai sekone, ma e mafai ona sui le mamanu e ala i le toe faasaʻoina o le faamatalaga upu. Na fausia le faiga e ala i le aoina ma le faailogaina o le faitau afe o ata fausiaga, ona faaleleia ai lea o se faataitaiga ata ua uma ona iai i le faaaogaina o le seti fou o faamaumauga. O le sini o lenei mea o le fesoasoani i tagata amata ina ia ausia se maualuga o le lelei e tutusa ma le tagata faapitoa.

Le Fa'ailogaina

E Uiga i le A' Design Award Uʻamea

O le A' Design Award Uʻamea na manumalo ai Technology Inheritance Meaʻai Autū: Meafaigaluega Fatuina o Mamanu Fausiaga, o se faamaoniga mamalu e tuuina atu i mamanu lelei e faataunuu ai manaʻoga faapolofesa ma pisinisi e faamoemoeina mai galuega ua mamanuina lelei. O ē manumalo i le Uʻamea e faaalia ai le tomai tele, le faapitoa maualuga, le malamalama o tagata poto ma le mafai fatufatuaʻi o ē na faia. E faamoemoeina Technology Inheritance Meaʻai Autū: Meafaigaluega Fatuina o Mamanu Fausiaga e avea ma mea ua matuā vivii, mataʻina ma pelepelea e aogā ma fou. O lenei mamanu na manumalo i le A' Design Award Uʻamea e tuufaatasia ai faiga sili o le pisinisi ma e iai uiga faatekonolosi mafai ua fesoʻotaʻi ma le poto ma le taleni. Ona o lona faatuatuaina, e tatau ai Technology Inheritance Meaʻai Autū: Meafaigaluega Fatuina o Mamanu Fausiaga ona faaaloalogia ma ua mamanuina e tuuina mai ai faaleleiga o le olaga e fesoasoani e faia ai e le lalolagi se nofoaga e sili atu ona lelei.

I Tua o le Mamanu

Pepa Fa'asalalau

Fa'amaoniaga, sui o le 'au ma fa'amatalaga fa'atekinisi na saunia e le tagata mamanu.

Uiga Fa'apitoa: This project trains an AI that professionally generates the appearance of buildings. The AI can generate buildings of different architectural styles according to the text entered by the user. The AI only takes a few seconds to generate the building's appearance, and users can recreate the design by changing the text description. That is, designers can obtain design concept proposals through AI cooperation and continuously optimize the design.
Fua & Fa'apipi'i: Generating building exteriors using a fine-tuned stable diffusion technique.
Fa'ailoga: Artificial intelligence, Human-computer cooperation, Architectural appearance generation, Text generation design, Quick modification, Architectural style control
Tekonolosi Gaosiga: The effect of using stable diffusion to produce architectural appearance could be better. The reason is that this mode needs to sufficiently understand the design skills of different designers, resulting in a messy appearance design and low quality. For this, fine-tuning the existing AI is essential. So this project reworked the data set, fine-tuned the critical stable diffusion, and made an AI to generate the specified architectural style to assist the junior designer in the design.
Lu'itau o le Mamanu: The challenge of this design is to build a professional architectural design AI for generating high-quality and stylized designs. This project collected and marked thousands of architectural appearance renderings of dozens of well-known architects, created a data set of architectural appearance, and then handed the data set to AI for fine-tuning learning. This project builds an AI to generate a given building's exterior style.
Su'esu'ega: Research Background: It is difficult for beginners to design high-quality architectural appearances.
Methods: This project trained an AI to professionally generate building exteriors, which can contract with designers to complete architectural exterior designs.
Results: Using the AI trained in this project, the designer can generate an architectural appearance rendering in just a few seconds.
Insights: Using this method, the architectural appearance of a specified style can be generated and modified through text description, and the quality of the generated design is close to that of a professional designer. This method does not require manual modeling and rendering, which improves design efficiency and quality.
Fa'aosofia: Most people measure the quality of a building by evaluating how it looks. Good architectural exterior design can bring attention and economic benefits. However, the exterior architectural design is complex, and the quality of architectural exteriors designed by excellent designers and beginners varies greatly. This design makes an artificial intelligence (AI) that can expertly generate building exteriors to help beginners achieve designs that approach the quality of master builders.
Umi o le Poloketi: The project has already been completed in 2023.
Fa'agasologa o Galuega: 1. Professional designers collect and mark, and screen images of outstanding designer building exteriors.
2. Use the filtered data to train an AI that specializes in generating the exterior of a given architectural style.
3. Designers use AI to design.
4. The designer inputs the design text description into the AI to specify the decoration style.
5. AI generates the interior design of the specified decoration style.
6. Modify the unsatisfactory details in the design through text description, and regenerate the design.
7. After adjusting the text description and generating it multiple times, a good design is obtained.