JAIST Repository >
School of Knowledge Science >
Articles >
Journal Articles >

Please use this identifier to cite or link to this item: http://hdl.handle.net/10119/19344

Title: Retrieval-Augmented Multi-Floor Building Image Generation
Authors: Wang, Zhengyang
Jin, Hao
Xie, Haoran
Keywords: Image generation
Retrieval-augmented generation
Diffusion model
Issue Date: 2024-06-22
Publisher: 情報処理学会
Magazine name: 研究報告コンピュータグラフィックスとビジュアル情報学(CG)
Volume: 2024-CG-194
Number: 3
Start page: 1
End page: 4
Abstract: Demand for generating building images from text prompts grows, despite recent advances in diffusion models greatly enhancing image quality. The current generative models struggle with controlling the number of floors. To this end, we propose a retrieval-augmented framework for generating building images with provided floor count using a diffusion model. Initially, the text prompts with the provided floor count to retrieve the most suitable image from a building image database. Then, we adopted a multi-level structure detection algorithm to obtain a sketch from the matched image to ensure structural consistency. Finally, the building image with the desired floor count and style is generated by diffusion model, guided by the detected building sketch. Our proposed framework enables accurate control over the floor count in building image synthesis. We demonstrate the robustness and scalability of generating building images with a specific floor count from text prompts.
Rights: 社団法人情報処理学会, Zhengyang Wang, Hao Jin, Haoran Xie, 情報処理学会研究報告. コンピュータグラフィックスとビジュアル情報学(CG), 2024-CG-194 (3), 2024, pp.1-4. ここに掲載した著作物の利用に関する注意: 本著作物の著作権は(社)情報処理学会に帰属します。本著作物は著作権者である情報処理学会の許可のもとに掲載するものです。ご利用に当たっては「著作権法」ならびに「情報処理学会倫理綱領」に従うことをお願いいたします。 Notice for the use of this material: The copyright of this material is retained by the Information Processing Society of Japan (IPSJ). This material is published on this web site with the agreement of the author (s) and the IPSJ. Please be complied with Copyright Law of Japan and the Code of Ethics of the IPSJ if any users wish to reproduce, make derivative work, distribute or make available to the public any part or whole thereof. All Rights Reserved, Copyright (C) Information Processing Society of Japan.
URI: http://hdl.handle.net/10119/19344
Material Type: publisher
Appears in Collections:a10-1. 雑誌掲載論文 (Journal Articles)

Files in This Item:

File Description SizeFormat
H-XIE-K-2.pdf337KbAdobe PDFView/Open

All items in DSpace are protected by copyright, with all rights reserved.

 


Contact : Library Information Section, Japan Advanced Institute of Science and Technology