Bag of tricks for real-world single image super-resolution

Yunfan Zhou

doi:10.1117/12.3039602

4 September 2024 Bag of tricks for real-world single image super-resolution

Yunfan Zhou

Proceedings Volume 13259, International Conference on Automation Control, Algorithm, and Intelligent Bionics (ACAIB 2024); 132591D (2024) https://doi.org/10.1117/12.3039602
Event: Fourth International Conference on Automation Control, Algorithm, and Intelligent Bionics (ICAIB 2024), 2024, Yinchuan, China

Abstract

Models built with deep learning has recently shown great success for addressing the challenging task of single image superresolution (SR). In-depth analyses of the representative models (e.g., EDSR, RCAN etc.) indicate the state-of-the-art performance is achieved by not only the claimed contributions (e.g., network structure, training strategies and loss functions) but also some underlying tricks, which is usually underemphasized. Our experimental analyses reveal that the tricks usually have a significant impact on the model’s performance, sometimes may play a more important role over the core designs. To gain more insight into this and provide a comprehensive analysis for providing SR related user guideline for those who are interested in SR, we’ve collected a bag of super-resolution tricks applicable to various stages (i.e., data pre-processing, data augmentation, loss function, training procedure, model structure, and model inference) of model implementation, and carried out extensive experiments to investigate what tricks are significant and how much can they improve the results over the base line. Through extensive evaluation, effect of different tricks is identified and thus encapsulated to various strong base lines for obtaining better quantitative and qualitative results. More importantly, the task of real-world SR benefit greatly from our formulated tricks, which is demonstrated to improve the PSNR by more than 0.4dB.

(2024) Published by SPIE. Downloading of the abstract is permitted for personal use only.

Citation Download Citation

Yunfan Zhou "Bag of tricks for real-world single image super-resolution", Proc. SPIE 13259, International Conference on Automation Control, Algorithm, and Intelligent Bionics (ACAIB 2024), 132591D (4 September 2024); https://doi.org/10.1117/12.3039602

ACCESS THE FULL ARTICLE

INSTITUTIONAL
Select your institution to access the SPIE Digital Library.

SELECT YOUR INSTITUTION

PERSONAL
Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.

PERSONAL SIGN IN

No SPIE Account? Create one

PURCHASE THIS CONTENT

SUBSCRIBE TO DIGITAL LIBRARY

50 downloads per 1-year subscription

Members: $195

Non-members: $335 ADD TO CART

25 downloads per 1 - year subscription

Members: $145

Non-members: $250 ADD TO CART

PURCHASE SINGLE ARTICLE

Includes PDF, HTML & Video, when available

Members: $17.00

Non-members: $21.00 ADD TO CART

PROCEEDINGS
11 PAGES

DOWNLOAD PAPER SAVE TO MY LIBRARY

GET CITATION

RIGHTS & PERMISSIONS

Get copyright permission Get copyright permission on Copyright Marketplace

KEYWORDS

Education and training

Data modeling

Performance modeling

RGB color model

Super resolution

Visual process modeling

Lawrencium

Show All Keywords

Keywords/Phrases

Search In:

Publication Years