{"id":13573632,"url":"https://github.com/CESNET/GPUJPEG","last_synced_at":"2025-04-04T12:31:22.411Z","repository":{"id":39613514,"uuid":"71791183","full_name":"CESNET/GPUJPEG","owner":"CESNET","description":"JPEG encoder and decoder library and console application for NVIDIA GPUs from CESNET and SITOLA of Faculty of Informatics at Masaryk University.","archived":false,"fork":false,"pushed_at":"2025-03-11T14:54:03.000Z","size":7993,"stargazers_count":252,"open_issues_count":7,"forks_count":71,"subscribers_count":20,"default_branch":"master","last_synced_at":"2025-03-11T15:49:31.530Z","etag":null,"topics":[],"latest_commit_sha":null,"homepage":"","language":"C","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":"bsd-2-clause","status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/CESNET.png","metadata":{"files":{"readme":"README.md","changelog":"ChangeLog","contributing":"CONTRIBUTING.md","funding":null,"license":"COPYING","code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null,"governance":null,"roadmap":null,"authors":"AUTHORS","dei":null,"publiccode":null,"codemeta":null}},"created_at":"2016-10-24T13:20:49.000Z","updated_at":"2025-03-11T14:53:56.000Z","dependencies_parsed_at":"2024-03-18T15:06:18.217Z","dependency_job_id":"0a5cdc44-e29a-452e-bbd9-b944fe1c8fcb","html_url":"https://github.com/CESNET/GPUJPEG","commit_stats":null,"previous_names":[],"tags_count":2,"template":false,"template_full_name":null,"repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/CESNET%2FGPUJPEG","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/CESNET%2FGPUJPEG/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/CESNET%2FGPUJPEG/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/CESNET%2FGPUJPEG/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/CESNET","download_url":"https://codeload.github.com/CESNET/GPUJPEG/tar.gz/refs/heads/master","host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":247179713,"owners_count":20897088,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2022-07-04T15:15:14.044Z","host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":[],"created_at":"2024-08-01T15:00:38.570Z","updated_at":"2025-04-04T12:31:22.399Z","avatar_url":"https://github.com/CESNET.png","language":"C","funding_links":[],"categories":["C"],"sub_categories":[],"readme":"GPUJPEG\n=======\n\n[![License](https://img.shields.io/badge/License-BSD_2--Clause-orange.svg)](https://opensource.org/licenses/BSD-2-Clause)\n[![C/C++ CI](../../workflows/C%2FC%2B%2B%20CI/badge.svg)](../../actions)\n\n![GPUJPEG Logo](logo.svg)\n\n**JPEG** encoder and decoder library and console application for **NVIDIA GPUs**\nfor high-performance image encoding and decoding. The software runs also on\n**AMD GPUs** using [ZLUDA](https://github.com/vosen/ZLUDA) (see\n[ZLUDA.md](ZLUDA.md)).\n\nThis documents provides an introduction to the library and how to use it. You\ncan also look to [FAQ.md](FAQ.md) for _performance tuning_\nand additional information. To see _latest changes_\nyou can display file [NEWS.md](NEWS.md).\n\nTable of contents\n-----------------\n- [Authors](#authors)\n- [Features](#features)\n- [Overview](#overview)\n- [Performance](#performance)\n  * [Encoding](#encoding)\n  * [Decoding](#decoding)\n- [Quality](#quality)\n- [Compile](#compile)\n- [Usage](#usage)\n  * [libgpujpeg library](#libgpujpeg-library)\n    + [Encoding](#encoding-1)\n    + [Decoding](#decoding-1)\n  * [GPUJPEG console application](#gpujpeg-console-application)\n- [Requirements](#requirements)\n- [License](#license)\n- [References](#references)\n\nAuthors\n-------\n - Martin Srom, CESNET z.s.p.o\n - Jan Brothánek\n - Petr Holub\n - Martin Jirman\n - Jiri Matela\n - Martin Pulec\n - Lukáš Ručka\n\nFeatures\n--------\n\n- uses NVIDIA CUDA platform\n- baseline Huffman 8-bit coding\n- use of **JFIF** file format by default, **Adobe** and **SPIFF** is supported as well (used by encoder\n  if JPEG internal color space is not representable by JFIF - eg. limited range **YCbCr BT.709** or **RGB**)\n- use of _restart markers_ that allow fast parallel encoding/decoding\n- Encoder by default creates _non-interleaved_ stream, optionally it can produce\n  an _interleaved_ stream (all components in one scan) or/and subsampled stream.\n- support for _color transformations_ and coding RGB JPEG\n- Decoder can decompress JPEG codestreams that can be generated by encoder. If scan\n  contains restart flags, decoder can use parallelism for fast decoding.\n-  _command-line_ tool with support for encoding/decoding **raw** images as well as **PNM/PAM** or **Y4M**\n\nOverview\n--------\n\nEncoding/Decoding of JPEG codestream is divided into following phases:\n\n     Encoding:                       Decoding\n     1) Input data loading           1) Input data loading\n     2) Preprocessing                2) Parsing codestream\n     3) Forward DCT                  3) Huffman decoder\n     4) Huffman encoder              4) Inverse DCT\n     5) Formatting codestream        5) Postprocessing\n\nand they are implemented on CPU or/and GPU as follows:\n - CPU:\n    - Input data loading\n    - Parsing codestream\n    - Huffman encoder/decoder (when restart flags are disabled)\n    - Output data formatting\n - GPU:\n    - Preprocessing/Postprocessing (color component parsing,\n      color transformation RGB \u003c-\u003e YCbCr)\n    - Forward/Inverse DCT (discrete cosine transform)\n    - Huffman encoder/decoder (when restart flags are enabled)\n\nPerformance\n-----------\n\nSource 16K (DCI) image ([8], [9]) was cropped to _15360x8640+0+0_ (1920x1080\nmultiplied by 8 in both dimensions) and for lower resolutions downscaled.\nEncoding was done with default values with input in RGB (quality **75**,\n**non-interleaved**, rst 24-36, average from 99 measurements excluding first\niteration) with following command:\n\n    gpujpegtool -v -e mediadivision_frame_\u003cres\u003e.pnm mediadivision_frame_\u003cres\u003e.jpg -n 100 [-q \u003cQ\u003e]\n\n### Encoding\n\n|      GPU \\ resolution      | HD (2 Mpix) | 4K (8 Mpix) | 8K (33 Mpix) | 16K (132 Mpix) |\n|----------------------------|-------------|-------------|--------------|----------------|\n|          RTX 3080          |   0.54 ms   |   1.71 ms   |    6.20 ms   |    24.48 ms    |\n|          RTX 2080 Ti       |   0.82 ms   |   2.89 ms   |   11.15 ms   |    46.23 ms    |\n|          GTX 1060M         |   1.36 ms   |   4.55 ms   |   17.34 ms   |  _(low mem)_   |\n|          GTX 580           |   2.38 ms   |   8.68 ms   |  _(low mem)_ |  _(low mem)_   |\n| AMD Radeon RX 7600 [ZLUDA] |   0.88 ms   |   3.16 ms   |   13.09 ms   |    50.52 ms    |\n\n**Note:** First iteration is slower because the initialization takes place and\nlasts about _28.6 ms_ for 8K (_87.1 ms_ for 16K) with RTX 3080 (but the\noverhead depends more on CPU than the GPU).\n\nFurther measurements were performed on _RTX 3080_ only:\n\n|             quality              | 10 | 20 | 30 | 40 | 50 | 60 | 70 | 80 | 90 | 100 |\n|----------------------------------|----|----|----|----|----|----|----|----|----|-----|\n| duration HD (ms)                 |0.48|0.49|0.50|0.51|0.51|0.53|0.54|0.57|0.60| 0.82|\n| duration 4K (ms)                 |1.61|1.65|1.66|1.67|1.69|1.68|1.70|1.72|1.79| 2.44|\n| duration 8K (ms)                 |6.02|6.04|6.09|6.14|6.12|6.17|6.21|6.24|6.47| 8.56|\n| duration 8K (ms, w/o PCIe xfers) |2.13|2.14|2.18|2.24|2.23|2.25|2.28|2.33|2.50| 5.01|\n\n\u003c!-- Additional notes (applies also for decode):\n 1. device needs to be set to maximum performance, otherwise powermanagement influences esp. PCIe transmits\n 2. stream formatter is starting to be a significant performance factor, eg. 0.82 ms for 8K Q=75 (contained in last line)\n 3. measurements were done without -DCMAKE_BUILD_TYPE=Release, should be measured with --\u003e\n\n### Decoding\n\nDecoded images were those encoded in previous section, averaging has been done similarly by\ntaking 99 samples excluding the first one. Command used:\n\n    gpujpegtool -v mediavision_frame_\u003cres\u003e.jpg output.pnm -n 100\n\n|       GPU \\ resolution     | HD (2 Mpix) | 4K (8 Mpix) | 8K (33 Mpix) | 16K (132 Mpix) |\n|----------------------------|-------------|-------------|--------------|----------------|\n|          RTX 3080          |   0.75 ms   |   1.94 ms   |    6.76 ms   |    31.50 ms    |\n|          RTX 2080 Ti       |   1.02 ms   |   1.07 ms   |   11.29 ms   |    44.42 ms    |\n|          GTX 1060M         |   1.68 ms   |   4.81 ms   |   17.56 ms   |  _(low mem)_   |\n|          GTX 580           |   2.61 ms   |   7.96 ms   | _(low mem)_  |  _(low mem)_   |\n| AMD Radeon RX 7600 [ZLUDA] |   1.00 ms   |   3.02 ms   |   11.25 ms   |    45.06 ms    |\n\n**Note**: _(low mem)_ above means that the card didn't have sufficient memory to\nencode or decode the picture.\n\nFollowing measurements were performed on _RTX 3080_ only:\n\n|              quality             | 10 | 20 | 30 | 40 | 50 | 60 | 70 | 80 | 90 | 100 |\n|----------------------------------|----|----|----|----|----|----|----|----|----|-----|\n| duration HD (ms)                 |0.58|0.60|0.63|0.65|0.67|0.69|0.73|0.78|0.89| 1.58|\n| duration 4K (ms)                 |1.77|1.80|1.83|1.84|1.87|1.89|1.92|1.95|2.11| 3.69|\n| duration 8K (ms)                 |6.85|6.88|6.90|6.92|6.98|6.70|6.74|6.84|7.17|12.43|\n| duration 8K (ms, w/o PCIe xfers) |2.14|2.18|2.21|2.24|2.27|2.29|2.34|2.42|2.71| 7.27|\n\nQuality\n-----------\nFollowing tables summarizes encoding quality and file size using NVIDIA\nGTX 580 for non-interleaved and non-subsampled stream with different quality\nsettings (PSNR and encoded size values are averages of encoding several\nimages, each of them multiple times):\n\n| quality |  PSNR 4K¹|   size 4K  |  PSNR HD²|  size HD   |\n|---------|----------|------------|----------|------------|\n|   10    | 29.33 dB |  539.30 kB | 27.41 dB |  145.90 kB |\n|   20    | 32.70 dB |  697.20 kB | 30.32 dB |  198.30 kB |\n|   30    | 34.63 dB |  850.60 kB | 31.92 dB |  243.60 kB |\n|   40    | 35.97 dB |  958.90 kB | 32.99 dB |  282.20 kB |\n|   50    | 36.94 dB | 1073.30 kB | 33.82 dB |  319.10 kB |\n|   60    | 37.96 dB | 1217.10 kB | 34.65 dB |  360.00 kB |\n|   70    | 39.22 dB | 1399.20 kB | 35.71 dB |  422.10 kB |\n|   80    | 40.67 dB | 1710.00 kB | 37.15 dB |  526.70 kB |\n|   90    | 42.83 dB | 2441.40 kB | 39.84 dB |  768.40 kB |\n|  100    | 47.09 dB | 7798.70 kB | 47.21 dB | 2499.60 kB |\n\n\u003cb\u003e\u003csup\u003e1,2\u003c/sup\u003e\u003c/b\u003e sizes 4096x2160 and 1920x1080\n\nCompile\n-------\nTo build console application check [Requirements](#requirements) and go\nto `gpujpeg` directory (where [README.md](README.md) and [COPYING](COPYING)\nfiles are placed) and run `cmake` command:\n\n    cmake -DCMAKE_BUILD_TYPE=Release -DCMAKE_CUDA_ARCHITECTURES=native -Bbuild .\n    cmake --build build --config Release\n\nIn Linux, you can also use **autotools** to create a build recipe for\nthe library and the application or a plain old _Makefile.bkp_. However,\n_cmake_ is recommended.\n\nUsage\n-----\n\n### libgpujpeg library\nTo build _libgpujpeg_ library check [Compile](#compile).\n\nTo use library in your project you have to include library to your\nsources and linked shared library object to your executable:\n\n    #include \u003clibgpujpeg/gpujpeg.h\u003e\n\nFor simple library usage examples you look into subdirectory [examples](examples).\n\n#### Encoding\nFor encoding by libgpujpeg library you have to declare two structures\nand set proper values to them. The first is definition of encoding/decoding\nparameters, and the second is structure with parameters of input image:\n\n    struct gpujpeg_parameters param = gpujpeg_default_parameters();\n    // you can adjust parameters:\n    param.quality = 80; // (default value is 75)\n\n    struct gpujpeg_image_parameters param_image = gpujpeg_default_image_parameter();\n    param_image.width = 1920;\n    param_image.height = 1080;\n    param_image.color_space = GPUJPEG_RGB; // input colorspace (GPUJPEG_RGB\n                                           // default), can be also\n                                           // eg. GPUJPEG_YCBCR_JPEG\n    param_image.pixel_format = GPUJPEG_444_U8_P012;\n    // or eg. GPUJPEG_U8 for grayscale\n    // (default value is GPUJPEG_444_U8_P012)\n\nIf you want to use subsampling in JPEG format call following function,\nthat will set default sampling factors (2x2 for Y, 1x1 for Cb and Cr):\n\n    // Use 4:2:0 subsampling\n    gpujpeg_parameters_chroma_subsampling(\u0026param, GPUJPEG_SUBSAMPLING_420);\n\nOr define sampling factors by hand:\n\n    // User custom sampling factors\n    gpujpeg_parameters_chroma_subsampling(\u0026param, MK_SUBSAMPLING(4, 4, 1, 2, 2, 1, 0, 0));\n\nNext you can initialize CUDA device by calling (if not called, default CUDA\ndevice will be used):\n\n    if ( gpujpeg_init_device(device_id, 0) )\n        return -1;\n\nwhere first parameters is CUDA device (e.g. `device_id = 0`) id and second\nparameter is flag if init should be verbose (`0` or `GPUJPEG_INIT_DEV_VERBOSE`).\nNext step is to create encoder:\n\n    struct gpujpeg_encoder* encoder = gpujpeg_encoder_create(0);\n    if ( encoder == NULL )\n        return -1;\n\nWhen creating encoder, library allocates all device buffers which will be\nneeded for image encoding and when you encode concrete image, they are\nalready allocated and encoder will used them for every image. Now we need\nraw image data that we can encode by encoder, for example we can load it\nfrom file:\n\n    size_t image_size = 0;\n    uint8_t* input_image = NULL;\n    if ( gpujpeg_image_load_from_file(\"input_image.rgb\", \u0026input_image,\n             \u0026image_size) != 0 )\n        return -1;\n\nNext step is to encode uncompressed image data to JPEG compressed data\nby encoder:\n\n    struct gpujpeg_encoder_input encoder_input;\n    gpujpeg_encoder_input_set_image(\u0026encoder_input, input_image);\n\n    uint8_t* image_compressed = NULL;\n    int image_compressed_size = 0;\n    if ( gpujpeg_encoder_encode(encoder, \u0026encoder_input, \u0026image_compressed,\n             \u0026image_compressed_size) != 0 )\n        return -1;\n\nCompressed data are placed in internal encoder buffer so we have to save\nthem somewhere else before we start encoding next image, for example we\ncan save them to file:\n\n    if ( gpujpeg_image_save_to_file(\"output_image.jpg\", image_compressed,\n             image_compressed_size, NULL) != 0 )\n        return -1;\n\nNow we can load, encode and save next image or finish and move to clean up\nencoder. Finally we have to clean up so destroy loaded image and destroy\nthe encoder.\n\n    gpujpeg_image_destroy(input_image);\n    gpujpeg_encoder_destroy(encoder);\n\n#### Decoding\nFor decoding we don't need to initialize two structures of parameters.\nWe only have to initialize CUDA device if we haven't initialized it yet and\ncreate decoder:\n\n    if ( gpujpeg_init_device(device_id, 0) )\n        return -1;\n    \n    struct gpujpeg_decoder* decoder = gpujpeg_decoder_create(0);\n    if ( decoder == NULL )\n        return -1;\n\nNow we have **two options**. The first is to do nothing and decoder will\npostpone buffer allocations to decoding first image where it determines\nproper image size and all other parameters (recommended).  The second option\nis to provide input image size and other parameters (reset interval, interleaving)\nand the decoder will allocate all buffers and it is fully ready when encoding\neven the first image:\n\n    // you can skip this code below and let the decoder initialize automatically\n    struct gpujpeg_parameters param;\n    gpujpeg_set_default_parameters(\u0026param);\n    param.restart_interval = 16;\n    param.interleaved = 1;\n    \n    struct gpujpeg_image_parameters param_image;\n    gpujpeg_image_set_default_parameters(\u0026param_image);\n    param_image.width = 1920;\n    param_image.height = 1080;\n    param_image.color_space = GPUJPEG_RGB;\n    param_image.pixel_format = GPUJPEG_444_U8_P012;\n    \n    // Pre initialize decoder before decoding\n    gpujpeg_decoder_init(decoder, \u0026param, \u0026param_image);\n\nIf you didn't initialize the decoder by `gpujpeg_decoder_init` but want\nto specify output image color space and subsampling factor, you can use\nfollowing code:\n\n    gpujpeg_decoder_set_output_format(decoder, GPUJPEG_RGB,\n                    GPUJPEG_444_U8_P012);\n    // or eg. GPUJPEG_YCBCR_JPEG and GPUJPEG_422_U8_P1020\n\nIf not called, RGB or grayscale is output depending on JPEG channel count.\n\nNext we have to load JPEG image data from file and decoded it to raw\nimage data:\n\n    size_t image_size = 0;\n    uint8_t* image = NULL;\n    if ( gpujpeg_image_load_from_file(\"input_image.jpg\", \u0026image,\n             \u0026image_size) != 0 )\n        return -1;\n    \n    struct gpujpeg_decoder_output decoder_output;\n    gpujpeg_decoder_output_set_default(\u0026decoder_output);\n    if ( gpujpeg_decoder_decode(decoder, image, image_size,\n             \u0026decoder_output) != 0 )\n        return -1;\n\nNow we can save decoded raw image data to file and perform cleanup:\n\n    if ( gpujpeg_image_save_to_file(\"output_image.pnm\", decoder_output.data,\n             decoder_output.data_size, \u0026decoder_output.param_image) != 0 )\n        return -1;\n    \n    gpujpeg_image_destroy(image);\n    gpujpeg_decoder_destroy(decoder);\n\n### GPUJPEG console application\nThe console application gpujpeg uses _libgpujpeg_ library to demonstrate\nit's functions. To build console application check [Compile](#compile).\n\nTo encode image from raw RGB image file to JPEG image file use following\ncommand:\n\n    gpujpegtool --encode --size=WIDTHxHEIGHT --quality=QUALITY \\\n            INPUT_IMAGE.rgb OUTPUT_IMAGE.jpg\n\nYou must specify input image size by `--size=WIDTHxHEIGHT` parameter.\nOptionally you can specify desired output quality by parameter\n`--quality=QUALITY` which accepts values 0-100. Console application accepts\na few more parameters and you can list them by folling command:\n\n    gpujpegtool --help\n\nTo decode image from JPEG image file to raw RGB image file use following\ncommand:\n\n    gpujpegtool --decode OUTPUT_IMAGE.jpg INPUT_IMAGE.rgb\n\nYou can also encode and decode image to test the console application:\n\n    gpujpegtool --encode --decode --size=WIDTHxHEIGHT --quality=QUALITY \\\n            INPUT_IMAGE.rgb OUTPUT_IMAGE.jpg\n\nDecoder will create new decoded file `OUTPUT_IMAGE.jpg.decoded.rgb` and do\nnot overwrite your `INPUT_IMAGE.rgb` file.\n\nConsole application is able to load raw RGB image file data from *.rgb\nfiles and raw YUV and YUV422 data from *.yuv files. For YUV422 you must\nspecify *.yuv file and use `--sampling-factor=4:2:2` parameter.\n\nAll supported parameters for console application are following:\n\n    --help\n        Prints console application help\n    --size=1920x1080\n        Input image size in pixels, e.g. 1920x1080\n    --pixel-format=444-u8-p012\n        Input/output image pixel format ('u8', '444-u8-p012', '444-u8-p012z',\n        '444-u8-p0p1p2', '422-u8-p1020', '422-u8-p0p1p2' or '420-u8-p0p1p2')\n    --colorspace=rgb\n        Input image colorspace (supported are 'rgb', 'yuv' and 'ycbcr-jpeg',\n        where 'yuv' means YCbCr ITU-R BT.601), when *.yuv file is specified,\n        instead of default 'rgb', automatically the colorspace 'yuv' is used\n    --quality\n        Set output quality level 0-100 (default 75)\n    --restart=8\n        Set restart interval for encoder, number of MCUs between\n        restart markers\n    --subsampled\n        Produce chroma subsampled JPEG stream\n    --interleaved\n        Produce interleaved stream\n    --encode\n        Encode images\n    --decode\n        Decode images\n    --device=0\n        By using this parameter you can specify CUDA device id which will\n        be used for encoding/decoding.\n\nRestart interval is important for parallel huffman encoding and decoding.\nWhen `--restart=N` is used (default is 8), the coder can process each\nN MCUs independently, and so he can code each N MCUs in parallel. When\n`--restart=0` is specified, restart interval is disabled and the coder\nmust use CPU version of huffman coder (because on GPU would run only one\nthread, which is very slow).\n\nThe console application can encode/decode multiple images by following\ncommand:\n\n    gpujpegtool ARGUMENTS INPUT_IMAGE_1.rgb OUTPUT_IMAGE_1.jpg \\\n            INPUT_IMAGE_2.rgb OUTPUT_IMAGE_2.jpg ...\n\nRequirements\n------------\nTo be able to build and run libgpujpeg **library** and gpujpeg **console\napplication** you need:\n\n1. NVIDIA **CUDA Toolkit**\n2. **C/C++ compiler** + **CMake**\n3. CUDA enabled **NVIDIA GPU** (cc \u003e= 2.0; older may or may not work) with\n   NVIDIA drivers _or_ AMD with [ZLUDA](https://github.com/vosen/ZLUDA)\n   (see [ZLUDA.md](ZLUDA.md))\n4. _optional_ **OpenGL** support:\n   - _GLEW_, _OpenGL_ (usually present in Windows, may need headers installation in Linux)\n   - _GLFW_ or _GLX_ (Linux only) for context creation\n   - _GLUT_ for OpenGL tests\n\nLicense\n-------\n- See file [COPYING](COPYING).\n- This software contains source code provided by NVIDIA Corporation.\n- This software source code is based on SiGenGPU \\[[3]\\].\n\nReferences\n----------\n[1]: http://www.w3.org/Graphics/JPEG/itu-t81.pdf\n[2]: http://www.ijg.org/\n[3]: https://github.com/silicongenome/SiGenGPU\n[4]: https://www.ecma-international.org/publications/files/ECMA-TR/ECMA%20TR-098.pdf\n[5]: https://www.itu.int/rec/dologin_pub.asp?lang=e\u0026id=T-REC-T.84-199607-I!!PDF-E\u0026type=items\n[6]: https://www.fileformat.info/format/spiff/egff.htm\n[7]: https://docs.oracle.com/javase/8/docs/api/javax/imageio/metadata/doc-files/jpeg_metadata.html#optcolor\n[8]: https://www.newsshooter.com/2019/07/25/beyond-imax-filming-with-a-gigantic-16k-200mp-sensor/\n[9]: https://we.tl/t-mjlrZM99EB\n\n1. [ITU-T Rec T.81][1]\n2. [ILG][2]\n3. [SiGenGPU][3] (currently defunct)\n4. [ECMA TR/098 (JFIF)][4]\n5. [ITU-T Rec T.84 (SPIFF)][5]\n6. [SPIFF File Format Summary (FileFormat.Info)][6]\n7. [Component ID registration][7]\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2FCESNET%2FGPUJPEG","html_url":"https://awesome.ecosyste.ms/projects/github.com%2FCESNET%2FGPUJPEG","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2FCESNET%2FGPUJPEG/lists"}