東華大學圖書館 |

Language: English

Help

回圖書館首頁

手機版館藏查詢

Back

Switch To: Labeled | MARC Mode | ISBD

Image Restoration, Editing, and Asse...

Ji, Liya,

Linked to FindBook

Google Book

Amazon

博客來

Image Restoration, Editing, and Assessment with Generative Artificial Intelligence /

Record Type:	Electronic resources : Monograph/item
Title/Author:	Image Restoration, Editing, and Assessment with Generative Artificial Intelligence // Liya Ji.
Author:	Ji, Liya,
Description:	1 electronic resource (106 pages)
Notes:	Source: Dissertations Abstracts International, Volume: 86-06, Section: A.
Contained By:	Dissertations Abstracts International86-06A.
Subject:	Language. -
Online resource:	https://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=31810371
ISBN:	9798346823551

Image Restoration, Editing, and Assessment with Generative Artificial Intelligence /
Ji, Liya,

Image Restoration, Editing, and Assessment with Generative Artificial Intelligence /Liya Ji. - 1 electronic resource (106 pages)

Source: Dissertations Abstracts International, Volume: 86-06, Section: A.

Modern high-performance computing (HPC) systems rely extensively on parallel architectures, including multicore CPUs and GPUs, to fulfill the ever-growing computational demands of today's applications. These architectures leverage multiple computational cores to execute tasks in parallel, thereby reducing execution time. However, as the core count continues to grow, so does the volume of data processed, exerting significant pressure on the underlying interconnection and memory systems. Additionally, the interaction between the processing and memory nodes makes it more complicated and challenging to design a high-performance and energy-efficient system. This thesis focuses on modeling, designing, and optimizing interconnection and memory systems for parallel architectures to enhance overall performance. Firstly, we systematically study the memory behavior of multithreaded applications running on shared-memory multicore CPUs. We propose a locality model named Shalom to characterize the memory behavior and enable fast and accurate statistical simulation of the applications. Secondly, we propose selective compression for network-on-chip (NoC), which dynamically compresses the packets based on the predicted compression ratios and the network status. We further present a NoC architecture with selective compression and power gating, SCNoCs, which achieves high performance while maintaining high energy efficiency. Lastly, we optimize the memory access in modern cluster-based GPU architectures by proposing collaborative coalescing, which effectively reduces the redundant memory access and alleviates NOC pressure, resulting in improved system performance.Firstly, we try to solve the task of inverse imaging problems via unconditional diffusion models. The inverse imaging problem aims to restore unknown clean images from input measurements that have incomplete information. We propose a learnable stateestimator-based diffusion model to incorporate the measurements into the reconstruction process and conduct experiments on typical inverse imaging problems (both linear and non-linear) with state-of-the-art results.Then, we extend the inverse imaging problem to instruction-based image editing. Unlike inverse imaging, filling the missing information with the prior of clean images, we want to edit the image with open-domain instruction guidance, challenging us with the higher requirements of scene understanding and generation. We approach the multimodality large language models to bridge understanding and generation for more complex cases. To achieve this goal, we break down the editing task with the multi-modality chain of thought prompts, i.e., task planning, editing region reasoning, and editing. Extensive experiments demonstrate that our method has competitive editing abilities on complex real-world images.Lastly, we switch the generation ability to the assessment aspect. With the emergence of AI-generated content on social media, image selection based on social preference is a rising and challenging task for the public. We gather Instagram's feedback to learn a popularity assessment model via a retrieval-augmented approach that uses both image features and user-specific statistics. Our approach achieves 79.5% accuracy, which significantly outperforms human and baseline models on the pairwise ranking of images from the Instagram Influencer Dataset. We also propose an application on the image selection on generative models, showing the effectiveness perceptually.

English

ISBN: 9798346823551Subjects--Topical Terms:

643551
Language.

Image Restoration, Editing, and Assessment with Generative Artificial Intelligence /
LDR:04792nmm a22003733i 4500 001 2430574
005 20250527055045.5
006 m o d
007 cr|nu||||||||
008 261011s2024 miu||||||m |||||||eng d
020 $a 9798346823551
035 $a (MiAaPQD)AAI31810371
035 $a AAI31810371
040 $a MiAaPQD $b eng $c MiAaPQD $e rda
100 1 $a Ji, Liya, $e author. $3 3818954
245 1 0 $a Image Restoration, Editing, and Assessment with Generative Artificial Intelligence / $c Liya Ji.
264 1 $a Ann Arbor : $b ProQuest Dissertations & Theses, $c 2024
300 $a 1 electronic resource (106 pages)
336 $a text $b txt $2 rdacontent
337 $a computer $b c $2 rdamedia
338 $a online resource $b cr $2 rdacarrier
500 $a Source: Dissertations Abstracts International, Volume: 86-06, Section: A.
500 $a Advisors: Chen, Qifeng; Yang, Qiang.
502 $b Ph.D. $c Hong Kong University of Science and Technology (Hong Kong) $d 2024.
520 $a Modern high-performance computing (HPC) systems rely extensively on parallel architectures, including multicore CPUs and GPUs, to fulfill the ever-growing computational demands of today's applications. These architectures leverage multiple computational cores to execute tasks in parallel, thereby reducing execution time. However, as the core count continues to grow, so does the volume of data processed, exerting significant pressure on the underlying interconnection and memory systems. Additionally, the interaction between the processing and memory nodes makes it more complicated and challenging to design a high-performance and energy-efficient system. This thesis focuses on modeling, designing, and optimizing interconnection and memory systems for parallel architectures to enhance overall performance. Firstly, we systematically study the memory behavior of multithreaded applications running on shared-memory multicore CPUs. We propose a locality model named Shalom to characterize the memory behavior and enable fast and accurate statistical simulation of the applications. Secondly, we propose selective compression for network-on-chip (NoC), which dynamically compresses the packets based on the predicted compression ratios and the network status. We further present a NoC architecture with selective compression and power gating, SCNoCs, which achieves high performance while maintaining high energy efficiency. Lastly, we optimize the memory access in modern cluster-based GPU architectures by proposing collaborative coalescing, which effectively reduces the redundant memory access and alleviates NOC pressure, resulting in improved system performance.Firstly, we try to solve the task of inverse imaging problems via unconditional diffusion models. The inverse imaging problem aims to restore unknown clean images from input measurements that have incomplete information. We propose a learnable stateestimator-based diffusion model to incorporate the measurements into the reconstruction process and conduct experiments on typical inverse imaging problems (both linear and non-linear) with state-of-the-art results.Then, we extend the inverse imaging problem to instruction-based image editing. Unlike inverse imaging, filling the missing information with the prior of clean images, we want to edit the image with open-domain instruction guidance, challenging us with the higher requirements of scene understanding and generation. We approach the multimodality large language models to bridge understanding and generation for more complex cases. To achieve this goal, we break down the editing task with the multi-modality chain of thought prompts, i.e., task planning, editing region reasoning, and editing. Extensive experiments demonstrate that our method has competitive editing abilities on complex real-world images.Lastly, we switch the generation ability to the assessment aspect. With the emergence of AI-generated content on social media, image selection based on social preference is a rising and challenging task for the public. We gather Instagram's feedback to learn a popularity assessment model via a retrieval-augmented approach that uses both image features and user-specific statistics. Our approach achieves 79.5% accuracy, which significantly outperforms human and baseline models on the pairwise ranking of images from the Instagram Influencer Dataset. We also propose an application on the image selection on generative models, showing the effectiveness perceptually.
546 $a English
590 $a School code: 1223
650 4 $a Language. $3 643551
650 4 $a Inverse problems. $3 3686950
650 4 $a Planning. $3 552734
650 4 $a Photography. $3 518339
650 4 $a Image retrieval. $3 3562846
650 4 $a Neural networks. $3 677449
650 4 $a Explicit knowledge. $3 3682762
650 4 $a Large language models. $3 3775019
650 4 $a Image. $2 rasuqam $3 1548898
650 4 $a Colorization. $3 3688181
690 $a 0679
690 $a 0338
710 2 $a Hong Kong University of Science and Technology (Hong Kong). $e degree granting institution. $3 3814653
720 1 $a Chen, Qifeng $e degree supervisor.
720 1 $a Yang, Qiang $e degree supervisor.
773 0 $t Dissertations Abstracts International $g 86-06A.
790 $a 1223
791 $a Ph.D.
792 $a 2024
856 4 0 $u https://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=31810371