Image super-resolution is one of the most popular computer vision problems
with many important applications to mobile devices. While many solutions have
been proposed for this task, they are usually not optimized even for common
smartphone AI hardware, not to mention more constrained smart TV platforms that
are often supporting INT8 inference only. To address this problem, we introduce
the first Mobile AI challenge, where the target is to develop an end-to-end
deep learning-based image super-resolution solutions that can demonstrate a
real-time performance on mobile or edge NPUs. For this, the participants were
provided with the DIV2K dataset and trained quantized models to do an efficient
3X image upscaling. The runtime of all models was evaluated on the Synaptics
VS680 Smart Home board with a dedicated NPU capable of accelerating quantized
neural networks. The proposed solutions are fully compatible with all major
mobile AI accelerators and are capable of reconstructing Full HD images under
40-60 ms while achieving high fidelity results. A detailed description of all
models developed in the challenge is provided in this paper.
Description
[2105.07825] Real-Time Quantized Image Super-Resolution on Mobile NPUs, Mobile AI 2021 Challenge: Report
%0 Generic
%1 ignatov2021realtime
%A Ignatov, Andrey
%A Timofte, Radu
%A Denna, Maurizio
%A Younes, Abdel
%A Lek, Andrew
%A Ayazoglu, Mustafa
%A Liu, Jie
%A Du, Zongcai
%A Guo, Jiaming
%A Zhou, Xueyi
%A Jia, Hao
%A Yan, Youliang
%A Zhang, Zexin
%A Chen, Yixin
%A Peng, Yunbo
%A Lin, Yue
%A Zhang, Xindong
%A Zeng, Hui
%A Zeng, Kun
%A Li, Peirong
%A Liu, Zhihuang
%A Xue, Shiqi
%A Wang, Shengpeng
%D 2021
%K CV_BokehSeminar
%T Real-Time Quantized Image Super-Resolution on Mobile NPUs, Mobile AI
2021 Challenge: Report
%U http://arxiv.org/abs/2105.07825
%X Image super-resolution is one of the most popular computer vision problems
with many important applications to mobile devices. While many solutions have
been proposed for this task, they are usually not optimized even for common
smartphone AI hardware, not to mention more constrained smart TV platforms that
are often supporting INT8 inference only. To address this problem, we introduce
the first Mobile AI challenge, where the target is to develop an end-to-end
deep learning-based image super-resolution solutions that can demonstrate a
real-time performance on mobile or edge NPUs. For this, the participants were
provided with the DIV2K dataset and trained quantized models to do an efficient
3X image upscaling. The runtime of all models was evaluated on the Synaptics
VS680 Smart Home board with a dedicated NPU capable of accelerating quantized
neural networks. The proposed solutions are fully compatible with all major
mobile AI accelerators and are capable of reconstructing Full HD images under
40-60 ms while achieving high fidelity results. A detailed description of all
models developed in the challenge is provided in this paper.
@misc{ignatov2021realtime,
abstract = {Image super-resolution is one of the most popular computer vision problems
with many important applications to mobile devices. While many solutions have
been proposed for this task, they are usually not optimized even for common
smartphone AI hardware, not to mention more constrained smart TV platforms that
are often supporting INT8 inference only. To address this problem, we introduce
the first Mobile AI challenge, where the target is to develop an end-to-end
deep learning-based image super-resolution solutions that can demonstrate a
real-time performance on mobile or edge NPUs. For this, the participants were
provided with the DIV2K dataset and trained quantized models to do an efficient
3X image upscaling. The runtime of all models was evaluated on the Synaptics
VS680 Smart Home board with a dedicated NPU capable of accelerating quantized
neural networks. The proposed solutions are fully compatible with all major
mobile AI accelerators and are capable of reconstructing Full HD images under
40-60 ms while achieving high fidelity results. A detailed description of all
models developed in the challenge is provided in this paper.},
added-at = {2022-12-17T14:24:52.000+0100},
author = {Ignatov, Andrey and Timofte, Radu and Denna, Maurizio and Younes, Abdel and Lek, Andrew and Ayazoglu, Mustafa and Liu, Jie and Du, Zongcai and Guo, Jiaming and Zhou, Xueyi and Jia, Hao and Yan, Youliang and Zhang, Zexin and Chen, Yixin and Peng, Yunbo and Lin, Yue and Zhang, Xindong and Zeng, Hui and Zeng, Kun and Li, Peirong and Liu, Zhihuang and Xue, Shiqi and Wang, Shengpeng},
biburl = {https://www.bibsonomy.org/bibtex/24834f7cf27f97c09a0567f14d2a39898/t_seizinger},
description = {[2105.07825] Real-Time Quantized Image Super-Resolution on Mobile NPUs, Mobile AI 2021 Challenge: Report},
interhash = {2a64dae55b1708e552239f3b4dbd253e},
intrahash = {4834f7cf27f97c09a0567f14d2a39898},
keywords = {CV_BokehSeminar},
note = {cite arxiv:2105.07825Comment: Mobile AI 2021 Workshop and Challenges: https://ai-benchmark.com/workshops/mai/2021/},
timestamp = {2022-12-17T14:24:52.000+0100},
title = {Real-Time Quantized Image Super-Resolution on Mobile NPUs, Mobile AI
2021 Challenge: Report},
url = {http://arxiv.org/abs/2105.07825},
year = 2021
}