Muhammad Usman Karim Khan - 2013-12-10

ces265 documentation

Introduction

Welcome to the documentatation of ces265, a "High Efficiency Video Coding" (HEVC) software written in C++ with multi-threading support. This software compresses the raw video stream in YUV 4:2:0 planar format into HEVC 8.0 compatible bitstream. Help is taken from HM-8.0, HM-9.2 and x265 softwares while writing this software. Refer to their original licenses before using this software. It is hoped that this software is useful but no guarantee of any sort is provided. The authors are not liable for any use or any damages caused by the use of this software.

There are several things to consider before using this software. First, there is no difference between a slice and a frame. In fact, only one slice per frame is allowed. Secondly, the frame dimensions must be divisible by the CTU_WIDTH and CTU_HEIGHT pre-processor directives (see Defines.h). Moreover, the CTU_WIDTH and CTU_HEIGHT cannot be made larger than 32. Currently, only Intra-encoding is supported in this software. However, the software is still developed to support Inter-encoding. Multi-threading is done via pthreads. This software is written and tested in Visual Studio 2012, and for pthreads, pthread-win32 is used. Though the software should also run fine on linux based platforms without any problems. Only tile-based threads are supported till now. However, the code is written in a manner to extend the threading capabilities to GOP- and slice-level as well. No deblocking filter is present.

Citation

In case of usage, please refer to our corresponding DATE 2014 publication:
M. U. Khan, M. Shafique, J. Henkel, "Software Architecture of High Efficiency Video Coding for Many-Core Systems with Power-Efficient Workload Balancing", IEEE/ACM Design Automation and Test in Europe (DATE), 2014.

Software Guide

User can provide input arguments via command line to this program. The command line arguments are given below with a brief detail. Note that the argument which is preceded by p is optional. The order in which the arguments appear is not important. For more information, see EncTop.cpp file.
1. "-i YUV420PFileName": The "-i" option is used to provide the yuv420 planar format file name to the program, which will be compressed by the program to generate the compressed bitstream as a file named "Video.h265".
2. "-w FrameWidth": The "-w" option specifies width of a frame in pixels.
3. "-h FrameHeight": The "-h" option specifies height of a frame in pixels.
4. p"-gop GopSize": The "-gop" option specifies the length of the GOP. The starting frame of a GOP is an Intra frame and all the rest are P-frames. Note that for the current implementation, GopSize must be equal to 1 as only Intra frame compression is supported. The default value of GopSize is equal to 1.
5. "-Nframes NumFrames": The "-Nframes" option specifies the total number of frames to compress.
6. p"-fps FramesPerSec": The "-fps" option is used to specify the frame-rate.
Note that this is only used while computing the RD-parameter and has no impact on compression or timing efficiency. The default value of FramesPerSec is 1.
7. p"-QP QPValue": The "-QP" options specifies the QP of all the frames. The default value of QPValue is 32.
8. p"-Ngopth NumGopThreads": The "-Ngopth" option specifies the total number of GOP threads used. For the current implementation, NumGopThreads must be equal to 1.
9. p"-Nsliceth NumSliceThreads": The "-Nsliceth" option specifies the total number of slice threads used. For the current implementation, NumSliceThreads must be equal to 1.
10. p"-Ntiles NumTilesPerFrame FrameWidthInTiles FrameHeightInTiles": The "-Ntiles" option specifies the total number of tiles that will reside in one full frame. Moreover, it also specifies the tile arrangement where FrameWidthInTiles argument gives the total tiles encompassing the width of the frame and FrameHeightInTiles argument does the same for the height of the frame. For example, "-Ntiles 20 5 4" will generate 20 tiles, 5 tile columns and 4 tile rows. For ces265, the sizes of the tiles are equal. Default value of NumTilesPerFrame is equal to 1.
11. p"-Ntileth NumTileThreads": The "-Ntileth" option specifies the total number of tile threads that will be used. The default value of NumTileThreads is 1.
12. p"--ver": The "--ver" option denotes verbosity and providing this argument to the program will produce verbose output. By default, verbosity is turned off.
13. p"--rec": The "--rec" option denotes reconstructed output generation. The name of the reconstructed yuv420 planar file is YUV420PFileName_HEVCRecon (see "-i" option). By default, no reconstructed output is generated.
14. p"--stat": The "--stat" option denotes writing output statistics in a "Statistics.txt" file. By default, no output statistics are written.

For an example, consider the following command line argument.

-i foreman_cif.yuv -w 352 -h 288 -Nframes 50 -Ntiles 2 2 1 -Ntileth 2 -fps 30 --rec --ver -QP 32 --stat

This will encode 50 frames of "foreman_cif.yuv" yuv420p video sequence, with dimensions 352x288 at a QP of 32, to generate "Video.h265" in the same directory as the program. Total of 2 tiles per frame are used, with 2 threads. Output is verbose. Reconstructed output "foreman_cif_HEVCRecon.yuv" will be generated in the same directory as the source yuv file. "Statistics.txt" will be generated in the same directory as the program. The fps of 30 is used to generate RD statistics in the "RD.txt" file present in the same directory as the program (see EncMain.cpp file).

Acknowledgement

This work was partly supported by the German Research Foundation (DFG) as part of the Transregional Collaborative Research Centre "Invasive Computing" (SFB/TR 89); http://invasic.de.

Credits

This software is written in "Chair for Embedded Systems (CES)", "Karlsruhe Institute of Technology (KIT)", Karlsruhe, Germany. Authors would like to thank the developers of HM and x265 software, and Vic Hargrave. Main contributors of this work
are:

 

Last edit: Muhammad Usman Karim Khan 2013-12-19