1 9 2 Virtual Reality Video Quality Assessment

1 9 2 Virtual Reality Video Quality Assessment Based on 3 D Convolutional Neural Networks Pei Wu, Wen. Xin Ding, Zhi. Xiang You, Ping An Shanghai University reporter：Pei Wu date： 2019/09/25 email: peiwu 1994@gmail. com

Introduction video-based model-based

Introduction Capture and stitch Projection transformation encode transmission play Back projection decode

Projection ERP（4096*2048） CMP 3*2（2880*1920） EAP（4096*2048） OHP（Octahedron 2880*1248）

database (a)Video 01(Talk) (d)Video 04(The Great Wall) (b)Video 02 (The Pagoda) (e)Video 05(Auditorium) (f)Video 06(Boat) (c)Video 03( Lake) (g)Video 07(concert)

database ERP EAP H. 264(QP) H. 265(QP) 30 30 35 35 40 40 45 45 50 50 7+7*2*2*5=147

Subjective quality assessment

Obiective quality assessment Kernel size Input channel Output channel Stride Padding Conv 1 2*5*5 3 32 1*2*2 0*2*2 Maxpool 1 3*2*2 32 32 3*2*2 0 Conv 2 2*5*5 32 64 1*2*2 0*2*2 Maxpool 2 2*2*2 64 64 2*2*2 0 Conv 3 1*5*5 64 128 1*2*2 2*2*0 Maxpool 3 1*4*4 128 1*4*4 0

Weight Calculation final score the objective fraction of all video patches in the VR video h h'

Result Method PLCC SRCC KRCC RMSE PSNR 0. 8485 0. 8680 0. 6589 0. 4774 WS-PSNR 0. 8937 0. 8956 0. 6834 0. 4348 SSIM 0. 9206 0. 9340 0. 7395 0. 3524 MSSIM 0. 9291 0. 9392 0. 7660 0. 3004 3 D-CNN 0. 9337 0. 9296 0. 7652 1. 1002 3 D-CNN（ERP） 0. 9414 0. 9601 0. 8632 1. 1265

1 9 2 THANK YOU FOR LISTENING reporter：Pei Wu email: peiwu 1994@gmail. com date： 2019/09/25