Intel® JPEG-XS Library

License

Scalable Video Technology is licensed under the OSI-approved BSD+Patent license. See LICENSE for details.

Notice

You are solely responsible for determining if your use of jpeg-xs requires any additional licenses. Intel is not responsible for obtaining any such licenses, nor liable for any licensing fees due, in connection with your use of jpeg-xs.

Overview

This library is implementation of ISO/IEC 21122 protocol.

Environment and Requirements

Encoder/Decoder sample app and libraries requirements:

• Any CPU that support x86-64 instruction set

Unit Tests requirements (SvtJpegxsUnitTests):

• Any CPU that support x86-64 and AVX2 instruction set

Supported OS versions:

• Linux Ubuntu 20.04 and 22.04
• Windows 10 and Windows 11

Build and Install

Windows* Operating Systems (64-bit)

• Build Requirements

  • Visual Studio* 2022 or 2019 (older version were not tested)
  • CMake 3.16 or later
  • YASM Assembler version 1.2.0 or later
    • Download the yasm executable from the following link
    • Rename yasm-*-win64.exe to yasm.exe
    • Copy yasm.exe into a location that is in the PATH environment variable (for example: %USERPROFILE%\AppData\Local\Microsoft\WindowsApps\)

• Build Instructions

  • Build the project by following the steps below
    • using terminal, cd into <repo dir>\Build\windows
    • run build.bat <2022|2019>, which will generate the Solution files (*.sln) and build the project

• Binaries and libraries Location

  • Binaries can be found under <repo dir>/Bin/Release or <repo dir>/Bin/Debug, depending on whether Debug or Release were selected in the build mode.

• API headers location

  • API headers can be found under Source/API

Linux* Operating Systems (64-bit)

• Build Requirements

  • GCC 9.4.0 or later
  • CMake 3.16 or later
  • YASM Assembler version 1.2.0 or later

• Build Instructions

  • cd Build/linux
  • ./build.sh <release | debug>

• Binaries and libraries location

  • Binaries can be found under Bin/Release and/or Bin/Debug

• API headers location

  • API headers can be found under Source/API

Encoder

Supported colour formats

PLANAR

format	EncApp: format + input-depth	ffmpeg name	status
YUV 420 8-bit	yuv420 + 8	yuv420p	Tested, working properly
YUV 420 10/12/14-bit	yuv420 + 10/12/14	yuv420p{10/12/14}le	Tested, working properly
YUV 422 8-bit	yuv422 + 8	yuv422p	Tested, working properly
YUV 422 10/12/14-bit	yuv422 + 10/12/14	yuv422p{10/12/14}le	Tested, working properly
YUV 444 8-bit	yuv444 + 8	yuv444p	Tested, working properly
YUV 444 10/12/14-bit	yuv444 + 10/12/14	yuv444p{10/12/14}le	Tested, working properly
RGB 8-bit	rgb + 8	gbrp	Tested, working properly
RGB 10/12/14-bit	rgb + 10/12/14	gbrp{10/12/14}le	Tested, working properly
YUV 400 8-bit	yuv400 + 8	-	Unsupported
YUV 400 10-bit	yuv400 + 10/12/14	-	Unsupported

PACKED

format	EncApp: format + input-depth	ffmpeg name	status
RGB 8bit	rgbp + 8	rgb24/bgr24	Tested, working properly, decoder decodes to planar format only
RGB 10/12/14bit	rgbp + 10/12/14	-	Tested, working properly, decoder decodes to planar format only

Usage examples

Windows

SvtJpegxsEncApp.exe -i <input_file.yuv> -b <output_bitstream.bin> -w 1920 -h 1080 --input-depth 8 --colour-format yuv422 --bpp 5 --decomp_v 2 --decomp_h 5 --lp 4

Linux

./SvtJpegxsEncApp -i <input_file.yuv> -b <output_bitstream.bin> -w 1920 -h 1080 --input-depth 8 --colour-format yuv422 --bpp 5 --decomp_v 2 --decomp_h 5 --lp 4

Compression ratio

To encode stream with compression rate of 12:1 (assuming 8bit yuv420p), one pixel require 12bits so --bpp 1 should be used:

./SvtJpegxsEncApp -i <input_file.yuv> -b <output_bitstream.bin> -w 1920 -h 1080 --input-depth 8 --colour-format yuv420 --bpp 1 --decomp_v 2 --decomp_h 5 --lp 4

Latency measurement

To measure average frame encoding time (latency) --limit-fps and high enough --lp parameters have to be used, cmd example:

./SvtJpegxsEncApp -i <input_file.yuv> -b <output_bitstream.bin> -w 1920 -h 1080 --input-depth 8 --colour-format yuv422 --bpp 5 --decomp_v 2 --decomp_h 5 --lp 4 --limit-fps 60

The --limit-fps must be lower than the maximum achievable number of frames. So that the next frames are not queued (time of waiting frames for processing is included in the latency time).

Interlaced video

To encode an interlaced stream, please provide a height that is half of the original. For example, for a 1080i stream, use -w 1920 -h 540 instead of -w 1920 -h 1080

Available parameters

Options:

[--help]                   Show usage options and exit
[--show-bands]             Print detailed informations about bands
                            (enabled:1, disable:0, default:0)
[-v]                       Verbose Encoder internal level (disabled:0, errors:1,
                            system info:2, system info full:3, warnings:4, ... all:6,
                            default: 2)

Input Options:

-i                         Input Filename
-w                         Frame width
-h                         Frame height
--colour-format            Set encoder colour format (yuv420, yuv422,  yuv444, rgb(planar), rgbp(packed))
                            (Experimental: yuv400)
--input-depth              Input depth
--bpp                      Bits Per Pixel, can be passed as integer or float
                            (example: 0.5, 3, 3.75, 5 etc.)
[-n]                       Number of frames to encode
[--limit-fps]              Limit number of frames per second
                            (disabled: 0, enabled [1-240])

Output Options:

[-b]                       Output filename
[--no-progress]            Do not print out progress (no print:1, print:0, default:0)
[--packetization-mode]     Specify how encoded stream is returned (multiple packets per frame:1, single packet per frame:0, default:0),
                           for details please refer to Codestream/Slice packetization mode
                           https://datatracker.ietf.org/doc/html/rfc9134

Coding features used during Rate Calculation (quality/speed tradeoff):

[--decomp_v]               Vertical decomposition (0, 1, 2, default: 2)
[--decomp_h]               Horizontal decomposition have to be greater or equal to
                            decomp_v (1, 2, 3, 4, 5, default: 5)
[--quantization]           Quantization method(deadzone:0, uniform:1, default:0)
[--slice-height]           The height of each slice in units of picture luma pixels (default:16, any value that is multiple of 2^(decomp_v))
[--coding-signs]           Enable Signs handling strategy (full:2, fast:1, disable:0, default:0)
[--coding-sigf]            Enable signification coding (enabled:1, disable:0, default:1)
[--coding-vpred]           Enable Vertical Prediction coding (disable:0, zero prediction residuals:1, zero coefficients:2, default: 0)
[--rc]                     Rate Control mode (
                            CBR: budget per precinct: 0,
                            CBR: budget per precinct with padding movement: 1,
                            CBR: budget per slice: 2,
                            CBR: budget per slice with nax size RATE: 3,
                            default 0)

Threading, performance:

[--asm]                    Limit assembly instruction set [0 - 11] or [c, mmx, sse, sse2,
                            sse3, ssse3, sse4_1, sse4_2, avx, avx2, avx512, max], by default
                            highest level supported by CPU
[--profile]                Threading model type, can be passed as digit (0, 1) or string
                            (latency, cpu) 0:latency (Low Latency mode), 1:cpu (Low CPU use
                            mode), default is 0
[--lp]                     Thread Scaling parameter, the higher the value the more threads
                            are created and thus lower latency and/or higher FPS can be
                            achieved (default: 0, which means lowest possible number of threads is created)

Decoder

Usage examples

Windows

SvtJpegxsDecApp.exe -i <input_bitstream.bin>  -o <output_file.yuv> --lp 5

Linux

./SvtJpegxsDecApp -i <input_bitstream.bin>  -o <output_file.yuv> --lp 5

Latency measurement

To measure average frame decoding time (latency) --limit-fps and high enough --lp parameters have to be used, cmd example:

./SvtJpegxsDecApp -i <input_bitstream.bin>  -o <output_file.yuv> --lp 5 --limit-fps 60

The --limit-fps must be lower than the maximum achievable number of frames. So that the next frames are not queued (time of waiting frames for processing is included in the latency time).

Available parameters

Options:

[--help]                   Show usage options and exit
[-v]                       Verbose decoder internal level (disabled:0, errors:1,
                            system info:2, system info full:3, warnings:4, ... all:6,
                            default: 2)
[--force-decode]           Force decode first frame

Input Options:

-i                         Input Filename
[-n]                       Number of frames to decode
[--find-bitstream-header]  Find bitstream header
[--limit-fps]              Limit number of frames per second
                            (disabled: 0, enabled [1-240])
[--packetization-mode]     Specify how bitstream is passed to decoder
                            (multiple packets per frame:1, single packet per frame:0, default:0)

Output Options:

[-o]                       Output Filename

Threading, performance:

[--asm]                    Limit assembly instruction set [0 - 11] or [c, mmx, sse, sse2, sse3,
                            ssse3, sse4_1, sse4_2, avx, avx2, avx512, max], by default highest
                            level supported by CPU
[--lp]                     Thread Scaling parameter, the higher the value the more threads are
                            created and thus lower latency and/or higher FPS can be achieved
                            (default: 0, which means lowest possible number of threads is created)

Encoder and Decoder design

Please see Encoder design

Please see Decoder design

API usage

Please see Encoder snippet for encoder structure overview and simplified encoder usage.

Please see Decoder snippet for decoder structure overview and simplified decoder usage.

Notes

The information in this document was compiled at v0.9 may not reflect the latest status of the design. For the most up-to-date settings and implementation, it's recommended to visit the section of the code.