Proceedings of the 2025 3rd International Conference on Image, Algorithms, and Artificial Intelligence (ICIAAI 2025)

AdaConv: Structure-Aware Real-Time Style Transfer Algorithm and Its Rendering Implementation in Unity Engine

Authors
Bin He1, *
1School of Digital Media, Shenzhen Institute of Information Technology, 2188 Longxiang Avenue, Longgang District, Shenzhen, Guangdong, 518172, China
*Corresponding author. Email: 2310160117@st.sziit.edu.cn
Corresponding Author
Bin He
Available Online 31 August 2025.
DOI
10.2991/978-94-6463-823-3_100How to use a DOI?
Keywords
AdaConv; Model Export; ONNX; Real-time Style Transfer; 3D Rendering
Abstract

This paper proposes a method for exporting the AdaConv style transfer algorithm to ONNX format and deploying it in Unity engine, addressing the structural compatibility issues between its dynamic kernel generation mechanism and static inference frameworks. This research designs a static conversion strategy based on grouped convolution vectorization, achieving an exportable fixed computation graph model while preserving structure-aware capabilities. The experiments export three structurally different model variants from the same AdaConv checkpoint: the Fully Dynamic Model maintains spatial and batch dimension dynamics, the Batch-Dynamic Model fixes spatial dimensions, and the Static Model adopts a completely static computation graph. All three models achieve real-time rendering performance of approximately 14.5 frames per second (fps) on the RTX 4060 platform, with SSIM metrics ranging approximately 0.67–0.68, supporting the mathematical equivalence of the static conversion. This approach applies to deploying structurally dynamic neural networks in inference environments requiring static graphs, providing an engineering feasibility solution for real-time three-dimensional style transfer applications.

Copyright
© 2025 The Author(s)
Open Access
Open Access This chapter is licensed under the terms of the Creative Commons Attribution-NonCommercial 4.0 International License (http://creativecommons.org/licenses/by-nc/4.0/), which permits any noncommercial use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license and indicate if changes were made.

Download article (PDF)

Volume Title
Proceedings of the 2025 3rd International Conference on Image, Algorithms, and Artificial Intelligence (ICIAAI 2025)
Series
Advances in Computer Science Research
Publication Date
31 August 2025
ISBN
978-94-6463-823-3
ISSN
2352-538X
DOI
10.2991/978-94-6463-823-3_100How to use a DOI?
Copyright
© 2025 The Author(s)
Open Access
Open Access This chapter is licensed under the terms of the Creative Commons Attribution-NonCommercial 4.0 International License (http://creativecommons.org/licenses/by-nc/4.0/), which permits any noncommercial use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license and indicate if changes were made.

Cite this article

TY  - CONF
AU  - Bin He
PY  - 2025
DA  - 2025/08/31
TI  - AdaConv: Structure-Aware Real-Time Style Transfer Algorithm and Its Rendering Implementation in Unity Engine
BT  - Proceedings of the 2025 3rd International Conference on Image, Algorithms, and Artificial Intelligence (ICIAAI 2025)
PB  - Atlantis Press
SP  - 1029
EP  - 1042
SN  - 2352-538X
UR  - https://doi.org/10.2991/978-94-6463-823-3_100
DO  - 10.2991/978-94-6463-823-3_100
ID  - He2025
ER  -