项目来源

日本学术振兴会基金(JSPS)

项目主持人

Phan・Xuan Tan

项目受资助机构

芝浦工業大学

立项年度

2024

立项时间

未公开

项目编号

24K20797

项目级别

国家级

研究期限

未知 / 未知

受资助金额

4420000.00日元

学科

知覚情報処理関連

学科代码

未公开

基金类别

若手研究

关键词

Light field coding ; Neural Radiance Fields ; View synthesis ; Scene rendering ; Light Field compression ; View Synthesis ; Depth Estimation ; Machine learning

参与者

未公开

参与机构

芝浦工業大学，工学部

项目标书摘要： per the original plan,we intended to implement the coding framework starting in AY2025.However,we successfully implemented it and obtained good results by the end of AY2024.We also extended the work to multiple scene compression.Interestingly,we started to build a real light field display,which was beyond the scope of the original plan.This project aims to represent light field scenes as implicit neural representations for compression,enabling them to be consumed like conventional images or videos.To achieve this goal,our main activities included:1)Surveying state-of-the-art techniques in light field compression,covering graph-based methods,learning-based approaches,and hybrid techniques;2)Studying monocular depth estimation and scene rendering technologies,which are expected to enhance the efficiency and quality of future compression frameworks;and 3)Implementing an implicit neural representation model capable of compressing(a)single light field scenes and(b)multiple light field scenes collectively.As a result of activities(1)and(2),we published research papers in an IEEE journal(IEEE Access)and a Springer international conference.Building upon(3),we have obtained promising experimental results,demonstrating high compression performance,and are preparing new submissions to journals and international conferences to further disseminate our findings.Building on the successful implementation of the implicit neural representation model for light field compression,we plan to extend the research in the following directions.-we will integrate and evaluate the compression system on the self-built light field display,aiming to achieve high rate-distortion performance,low-latency,high-quality visualization.-we will investigate adaptive coding methods and dynamic view synthesis to enhance user experiences.Based on these developments,we aim to publish in journals and international conferences.Reason:As per the original plan,we intended to implement the coding framework starting in AY2025.However,we successfully implemented it and obtained good results by the end of AY2024.We also extended the work to multiple scene compression.Interestingly,we started to build a real light field display,which was beyond the scope of the original plan。Outline of Research at the Start:Light Field imaging potentially provides more immersive and closer to reality multimedia experience to end-users for various applications.However,it introduces huge amount of high-dimensional data.This research aim at designing a novel coding scheme to efficiently compress Light Field data。