P‐45: Performance Optimization Methods for Large Image Generation Models on Low‐Computing‐Power Devices
SID Symposium Digest of Technical Papers · 2026-06 · 57 (1) · 1845-1848
DOI: 10.1002/sdtp.70001_501 ↗Kurz gefasst
Fachartikel in der Zeitschrift SID Symposium Digest of Technical Papers, online erschienen am 11.09.2026. Verfasst von Jiaojiao Meng und 6 weiteren Autor:innen. Die Arbeit nennt 11 Literaturangaben.
Abstract (Originalsprache)
This paper presents a system‐level optimization framework for large image generation models on low‐computing‐power devices, proposing the BCache temporal caching algorithm combined with resource‐adaptive memory management and computational graph optimization. This approach increases inference speed by 1.5‐4.13x while reducing peak GPU memory usage by 45.8%, all while maintaining nearly lossless generation quality. It provides a practical solution for efficient deployment of large image generation models on resource‐constrained devices.
Zitieren
APA
Meng, J., Li, J., Huang, X., Zu, C., Liu, H., Zhao, X., & Jiang, X. (2026). P‐45: Performance Optimization Methods for Large Image Generation Models on Low‐Computing‐Power Devices. SID Symposium Digest of Technical Papers, 57(1), 1845-1848. https://doi.org/10.1002/sdtp.70001_501
Vancouver
Meng J, Li J, Huang X, Zu C, Liu H, Zhao X, et al. P‐45: Performance Optimization Methods for Large Image Generation Models on Low‐Computing‐Power Devices. SID Symposium Digest of Technical Papers. 2026;57(1):1845-1848. doi:10.1002/sdtp.70001_501
BibTeX
@article{Meng2026,
author = {Meng, Jiaojiao and Li, Jianjie and Huang, Xile and Zu, Chunshan and Liu, Hanwen and Zhao, Xingxing and Jiang, Xingqun},
title = {P‐45: Performance Optimization Methods for Large Image Generation Models on Low‐Computing‐Power Devices},
journal = {SID Symposium Digest of Technical Papers},
year = {2026},
volume = {57},
number = {1},
pages = {1845--1848},
doi = {10.1002/sdtp.70001_501}
}
EUVIMED