RAPIQUE

An official implementation of Rapid and Accurate Video Quality Evaluator (RAPIQUE) proposed in [IEEE OJSP2021] RAPIQUE: Rapid and Accurate Video Quality Prediction of User Generated Content. Arxiv. IEEExplore(Open Access!) and [PCS2021] Efficient User-Generated Video Quality Prediction. IEEExplore. Note that the temporal features can be used as standalone features in company with spatial models to boost performance on motion-intensive models. Check out the temporal-only modules in [ICIP21] A Temporal Statistics Model For UGC Video Quality Prediction. IEEExplore

Check out our BVQA resource list and performance benchmark/leaderboard results in https://github.com/vztu/BVQA_Benchmark.

For more evaluation codes, please check out VIDEVAL

Requirements

MATLAB >= 2019
- Deep learning toolbox (ResNet-50)
Python3
Sklearn
FFmpeg
Git LFS

Performances

SRCC / PLCC

Methods	KoNViD-1k	LIVE-VQC	YouTube-UGC	All-Combined
TLVQM	0.7101 / 0.7037	0.7988 / 0.8025	0.6693 / 0.6590	0.7271 / 0.7342
VIDEVAL	0.7832 / 0.7803	0.7522 / 0.7514	0.7787 / 0.7733	0.7960 / 0.7939
MDVSFA	0.7812 / 0.7856	0.7382 / 0.7728	- / -	- / -
RAPIQUE	0.8031 / 0.8175	0.7548 / 0.7863	0.7591 / 0.7684	0.8070 / 0.8229

Scatter plots and fitted logistic curves on these datasets:

KonVid-1k	LIVE-VQC	YouTube-UGC	All-Combined

Speed

The unit is average secs/video.

Methods	540p	720p	1080p	4k@60
Video-BLIINDS	341.1	839.1	1989.9	16129.2
VIDEVAL	61.9	146.5	354.5	1716.3
TLVQM	34.5	78.9	183.8	969.3
RAPIQUE	13.5	17.3	18.3	112

Performance vs. Speed

Demos

Feature Extraction Only

demo_compute_RAPIQUE_feats.m

You need to specify the parameters

Evaluation of BVQA Model

We proposed several evaluation methods for BIQA/BVQA models. Please check out [ICASSP21] Regression or classification? New methods to evaluate no-reference picture and video quality models IEEExplore for details.

For regression evaluation:

$ python evaluate_bvqa_features_regression.py

For binary classification evaluation:

$ python evaluate_bvqa_features_binary_classification.py

For ordinal classification evaluation:

$ python evaluate_bvqa_features_ordinal_classification.py

Citation

If you use this code for your research, please cite our papers.

@article{tu2021rapique,
  title={RAPIQUE: Rapid and accurate video quality prediction of user generated content},
  author={Tu, Zhengzhong and Yu, Xiangxu and Wang, Yilin and Birkbeck, Neil and Adsumilli, Balu and Bovik, Alan C},
  journal={IEEE Open Journal of Signal Processing},
  volume={2},
  pages={425--440},
  year={2021},
  publisher={IEEE}
}
@article{tu2021ugc,
  title={UGC-VQA: Benchmarking blind video quality assessment for user generated content},
  author={Tu, Zhengzhong and Wang, Yilin and Birkbeck, Neil and Adsumilli, Balu and Bovik, Alan C},
  journal={IEEE Transactions on Image Processing},
  year={2021},
  publisher={IEEE}
}
@inproceedings{tu2021efficient,
  title={Efficient User-Generated Video Quality Prediction},
  author={Tu, Zhengzhong and Chen, Chia-Ju and Wang, Yilin and Birkbeck, Neil and Adsumilli, Balu and Bovik, Alan C},
  booktitle={2021 Picture Coding Symposium (PCS)},
  pages={1--5},
  year={2021},
  organization={IEEE}
}
@inproceedings{tu2021temporal,
  title={A Temporal Statistics Model For UGC Video Quality Prediction},
  author={Tu, Zhengzhong and Chen, Chia-Ju and Wang, Yilin and Birkbeck, Neil and Adsumilli, Balu and Bovik, Alan C},
  booktitle={2021 IEEE International Conference on Image Processing (ICIP)},
  pages={1454--1458},
  year={2021},
  organization={IEEE}
}
@inproceedings{tu2021regression,
  title={Regression or classification? New methods to evaluate no-reference picture and video quality models},
  author={Tu, Zhengzhong and Chen, Chia-Ju and Chen, Li-Heng and Wang, Yilin and Birkbeck, Neil and Adsumilli, Balu and Bovik, Alan C},
  booktitle={ICASSP 2021-2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)},
  pages={2085--2089},
  year={2021},
  organization={IEEE}
}

vztu / RAPIQUE