网站分析>> AVT-VQDB-UHD-1>> 返回
项目作者： Telecommunication-Telemedia-Assessment

项目描述 ：
4K database from AVT group of TU Ilmenau
高级语言： Python
项目地址: git://github.com/Telecommunication-Telemedia-Assessment/AVT-VQDB-UHD-1.git
创建时间： 2019-10-30T13:19:07Z
项目社区：https://github.com/Telecommunication-Telemedia-Assessment/AVT-VQDB-UHD-1
开源协议：GNU General Public License v3.0
下载


AVT-VQDB-UHD-1

This is a repository containing 4K databases that we initially published at IEEE-ISM 2019.

You can explore the dataset using the overview page.

However, for download it is recommended to use the provided tool, because the videos are not hosted in this repository.

News:

  • 01.03.2022: The database was updated with a new dataset related to codec retraining, see the folder codec_retraining.

Contents:

Acknowledgements

If you use any of the data or code, please cite the following paper:

  1. @inproceedings{rao2019Db,
  2.     author = {Rakesh Rao Ramachandra Rao and Steve G\"oring and Werner Robitza and Bernhard Feiten and Alexander Raake},
  3.     title = {AVT-VQDB-UHD-1: A Large Scale Video Quality Database for UHD-1},
  4.     booktitle={2019 IEEE ISM},
  5.     year = {2019},
  6.     pages={1-8},
  7.     volume={},
  8.     month={Dec}
  9. }

Download of source videos and segments

To download all video segments and source videos (in total around 55GB), use the provided download tool, e.g. a ./download.py --help call will show you the following help:

  1. usage: download.py [-h] [--no_sources] [--output OUTPUT]
  3. download video segments and source videos
  5. optional arguments:
  6.   -h, --help       show this help message and exit
  7.   --no_sources     download no sources (default: False)
  8.   --output OUTPUT  specify a different output directory (default: working directory)

You can also use your favorite download tool, check the base_url defined in download.py.

Structure

We shortly describe the folder structure used in this repository and after downloading of all additional files.

  • src_videos: source videos for all tests
  • test_x.json: a list of encoded video segments for each test
  • src_videos.json: a list of source videos

For each test:

  • test_X/mos_ci.csv: subjective ratings with CI values
  • test_X/crowd/mos_ci.csv: subjective ratings with CI values for the crowd test (only available for test_1)
  • test_X/metadata.csv: video meta-data for all video segments
  • test_X/objective_scores.csv: calculated objective video quality scores
  • test_X/segments: all encoded videos used in the test

Additional quality metric data:

  • test_X/brisque_reports: extracted features for BRISQUE model
  • test_X/niqe_reports: extracted features for NIQE model
  • test_X/vmaf_reports: all per-frame VMAF and other full ref scores

SITI-related:

  • siti/*.csv: SI/TI scores for each SRC video, according to ITU-T Rec. P.910, calculated using siti

Codec retraining:

  • codec_retraining/data.csv: Dataset for retraining P.1203.1 for other codecs

AVPVS generation

The described tests were coducted using AVPVS (audiovisual processed video sequences, a term used within ITU-T P.1203 and P.1204-related projects).

AVPVS are rescaled (to 3840 x 2160 @ 60fps matching the native resolution and framerate of the used display) and uncompressed versions of the encoded video segments, this ensures that there are no player specific influences while playing out the videos.
For the purpose of reducing storage, we only include the encoded videos in this Database, however the AVPVS version can be generated using the provided tool.

To run, e.g., full reference models, it is also required to perform such conversion to match the same resolution and framerate of the distorted video to the corresponding source video.
For this you can use the script gen_avpvs.py.
As a requirement you need Python >=3.6 installed.

For the usage, run ./gen_avpvs.py --help:

  1. usage: gen_avpvs.py [-h] [--avpvs_folder AVPVS_FOLDER] videosegment src
  3. convert video segment to avpvs
  5. positional arguments:
  6.   videosegment          video segment
  7.   src                   source video
  9. optional arguments:
  10.   -h, --help            show this help message and exit
  11.   --avpvs_folder AVPVS_FOLDER
  12.                         folder for storing the avpvs file (default: ./avpvs)

If you run the avpvs script, please be careful with the arguments videosegment and src, so that they are not swapped.

Papers using the Dataset

This dataset is used in the following papers:

  • Werner Robitza, Rakesh Rao Ramachandra-Rao, Steve Göring, Alexander Dethof, and Alexander Raake. 2022. “Deploying the ITU-T P.1203 QoE Model in the Wild and Retraining for New Codecs.” Mile-High Video Conference (MHV ’22), Denver, USA. March 2022. [url]
  • Steve Göring, Rakesh Rao Ramachandra Rao, Bernhard Feiten, and Alexander Raake. “Modular Framework and Instances of Pixel-based Video Quality Models for UHD-1/4K.” IEEE Access. vol. 9. 2021 [url]
  • Alexander Raake, Silvio Borer, Shahid Satti, Jörgen Gustafsson, Rakesh Rao Ramachandra Rao, Stefano Medagli, Peter List, Steve Göring, David Lindero, Werner Robitza, Gunnar Heikkilä, Simon Broom, Christian Schmidmer, Bernhard Feiten, Ulf Wüstenhagen, Thomas Wittmann, Matthias Obermann, and Roland Bitto. “Multi-model standard for bitstream-, pixel-based and hybrid video quality assessment of UHD/4K: ITU-T P.1204.” IEEE Access. vol. 8. 2020 [url]
  • Rakesh Rao Ramachandra Rao, Steve Göring, Werner Robitza, Alexander Raake, Bernhard Feiten, Peter List, and Ulf Wüstenhagen. “Bitstream-based Model Standard for 4K/UHD: ITU-T P.1204.3 — Model Details, Evaluation, Analysis and Open Source Implementation.” Twelfth International Conference on Quality of Multimedia Experience (QoMEX). Athlone, Ireland. May 2020. [url]
  • Steve Göring, Christopher Krämmer, and Alexander Raake. “cencro — Speedup of Video Quality Calculation using Center Cropping.” 21st IEEE International Symposium on Multimedia (2019 IEEE ISM). Dec 2019. [url]

Licenses

License for the code

GNU General Public License v3. See LICENSE file in this repository.

License for the shared videos

This database consists of short term videos based on several short movies, that are either public available or created by TU Ilmenau.
The tools provided in this repository can be used to download the shared videos that are used in the described video quality tests.
In the following we specify the common filename prefix, to identify the source and corresponding license of the video.
This applies to encoded and source videos that are shared within this database.
For example all files that can be downloaded with the prefix bigbuck_bunny are based on the Big Bucks Bunny content and follow the corresponding license.

We are happy that it was possible to access and use all the external video sources.

Our own contents follows the Attribution-NonCommercial 4.0 International (CC BY-NC 4.0) license.

  • Dancers_8s
  • cutting_orange_tuil
  • fr-041_debris
  • Daydreamer
  • vegetables
  • Giftmord