Video Stream Completion with real-time processing

Citation Author(s):
Lunji
Song
Lanzhou University
Guoxian
Zhu
Lanzhou University
Submitted by:
Lunji Song
Last updated:
Wed, 10/25/2023 - 22:58
DOI:
10.21227/53tc-tw58
Data Format:
License:
0
0 ratings - Please login to submit your rating.

Abstract 

The following videoes use the video streaming completion model, which combines static and dynamic information for real-time processing. The proposed model is solved using the alternating direction method of multipliers (ADMM), and using MATLAB for solution recovery.

Gray video suzie: This video is restored in the case of the missing rates set to 70%, 80%, 90%, respectively

Color Video Hall: This video is restored in the case of the missing rates set to 70%, 80%, and 90%, respectively

Color Video Flower: This video is recovered separately in the case of the missing rates set to 90%

Color video tempete: This video is restored separately in the case of the missing rates set to 90%

 

We evaluate our proposed model using two commonly used public tensor datasets: the gray-scale video dataset and the color video dataset (http://trace.eas.asu.edu/yuv/). These datasets are frequently used to assess the tensor completion performance of different models. We conducted tests on the gray video dataset named "suzie" with the size of 144$\times$176$\times$150. The missing rates were set to 80$\%$, 90$\%$, and 95$\%$, and the error tolerance $tol$ was set to [$10^{-6}$, $10^{-4}$]. We also conducted experiments on three color video datasets: $"Hall"$, "$Flower$"  and "$tempete$". The dimensions of these datasets are 144$\times$ 176 $\times$ 3 $\times$ 300. The initial size of the tensor is $\mathcal{X}^{1}(\in \mathbb{R}^{144\times 176 \times 3 \times d})$, where the parameter $d$ can be chosen based on the RSE values.

Instructions: 

The file “original_video name” represents the original image of the "video name"; "video name+miss90" represents the observed "video name" with a missing rate of 90%; "video name_completion90" shows the recovery result of our model for the  "video name"  with a missing rate 90%.

 

Funding Agency: 
National natural science foundations of China
Grant Number: 
12171216

Comments

A new model for "VIDEO STREAM COMPLETION WITH REAL-TIME PROCESSING".

Submitted by Lunji Song on Mon, 10/23/2023 - 10:56