HMK-CTA: A Hierarchical Multidimensional Representation for Visual Datasets

Yu-Ting Tsai, Muhammad Abu Bakar

Letter Of Changes:

We would like to thank the area chair and reviewers for detailed, insightful reviews, meta-review, and recommendations to improve the quality of this article. We have revised the manuscript according to reviewers’ comments and suggestions. In the following, we list major changes in this revision and answer reviewers’ comments/questions.

Major Changes

1. Add Eq. 2 and modify the introduction to MK-CTA to describe it in more detail. (Sec. 3.2)

2. For TVLFs and TVVD, add the reconstructed images of more time steps (Figs. 8 and 9)

3. Add the “Acknowledgements” section.

4. Fix some typos.

Responses to Reviewer TXNn (“C” stands for comments and “R” for response)

[C1]: For the time-varying data, it would be better to show results on a few time steps.

[R1]: Thank you for this great comment. In Figs 8 and 9, we added the reconstructed (and also enlarged) images of two additional time steps respectively for the TVLF “Animated Bunnies” and the TVVD “TurbJet”.

[C2]: Some typo errors should be fixed, e.g., ref [20], photrealistic -> photorealistic

[R2]: We fixed the mentioned typo and others in this revision.

Responses to Reviewer PqoE

[C1]: I think Sec. 3.2 needs more details to make readers clearly understand MK-CTA..

[R1]: Thank you for this great comment. In Sec 3.2, we added Eq. 2 to clearly explain how to reconstruct the original tensor based on the decomposed results of MK-CTA. We also modified the introduction to MK-CTA to describe it in more detail.

Keywords: Realtime Rendering, Multidimensional Data Analysis, Hierarchical Model, Multiway Clustering, Sparse Representation.

Abstract: This paper presents a novel tensor-based representation, namely hierarchical multiway K-clustered tensor approximation, for multidimensional visual datasets. The proposed method extends a previous tensor model into a hierarchical representation that can significantly reduce offline computational cost as well as provide similar approximation quality and rendering performance at the same time. We also apply the proposed method to approximate spatially-varying bidirectional reflectance distribution functions, time-varying light fields, and time-varying volume data to show its effectiveness and potential for data-driven rendering. Experimental results demonstrate that under similar performance to previous work, the proposed method can reduce offline approximation time by even an order of magnitude.

Submission Number: 15