1 Yossi Cohen Video Transcodin g & Transratin g
Jan 15, 2015
1
Yossi Cohen
Video Transcoding & Transrating
2
Overview
Full TranscodeOpen Loop transratersCoefficient Drop TransraterRe-quantization transrater
Closed loop transraterFPDT
3
Full TranscoderFull transcoding does full encoding and
decoding of the video. Features include:Easy ImplementationUsually used for
format conversion such as MPEG2 Video->H264Video Resize and frame rate reduction
Highest CPU requirements
4
Aided TranscoderFull decoding of the videoEncoding Video while re-using the decoded video
information including:Qp selectionMotion vectorsBlock size (8x8, 4x8 etc.) selection
MV, Qp
5
Aided Transcoder Features
High CPU requirements Like Full transcoder, enables:
Resizing Change of Frame type (P->I, I->P) for sync point changing Smooth frame rate conversion Accurate (no drift) transcoding
6
Open Loop Transcoders
Open loop transcoders are fast and simple transrater.
Open-loop – does not correct the drift error created by the transrating on next frames
Introduces drift error into the video – video quality seriously degrades as a function of the distance from last I frame.
7
Coefficient Drop Transrater
Performs only Entropy Decoding/EncodingDrops some of the coefficients according to
psycho-visual models.Features
Fastest transraterPoorest video quality per bitrate Introduce drift
8
Re-Quanitzation transraterPerforms Entropy decoding and de-
quantization followed by quantization with an Higher Qp (lowers quality & BR)
FeaturesFast TransraterPoor video quality per bitrate Introduce drift
9
Closed loop transrater (FPDT)Performs Entropy decoding and de-
quantization followed by quantization with an Higher Qp (lowers quality & BR)
Adds quantization error from previous frame to next frame to avoid drift
Error Store
10
Closed Loop Transrater - Features Faster then TranscodersNo driftQuality is transcoder grade per same bitrate
11
Quality / CPU comparisonLine – Quality, Bar – CPU Cycles
Full T
ranscode
Aided T
ranscoder
FP
DT Re-
Qua
ntization
transrater
Coefficien
t D
rop
12
Frame dropping overview
Frame dropping could improve KPI (5-10%) and reduce CPU load.
Smooth frame drop – convert to any fps, fixed deltaT between frames, frame interpolation. Very high CPU requirements
Regular frame drop – drop any frame introduce some jerkiness
Limited fps – cant drop any frame, introduce higher jerkiness
13
Frame Dropping -2
Transcoders support both Regular frame drop
Transraters only supports limited fps frame drop
FPDT transrater might support regular frame drop at additional development time
14
Frame Resize
Frame Resize could considerably lower bitrate.
CPU - Requires per pixel operations for resizing – higher CPU but lower encoded frame size which lowers CPU.
Transcoder could easily support resize – additional CPU requirements
FPDT could support resize but would cost considerable CPU and development time
15
Thank you!
Video Expert
Lectures on Video / Android / VoIP
Android Native Developer
More About me:
Yossi CohenYossi [email protected]://www.mobilevideotech.com
+972-545-313092+972-545-313092