CONVIQT: Contrastive Video Quality Estimator

Pavan C. Madhusudana, Neil Birkbeck, Yilin Wang, Balu Adsumilli and Alan C. Bovik

This is the official repository of the paper CONVIQT: Contrastive Video Quality Estimator

Usage

The code has been tested on Linux systems with python 3.7. Please refer to requirements.txt for installing dependent packages.

Running CONVIQT

In order to obtain quality score, checkpoints needs to be downloaded. The following command can be used to download the checkpoint.

wget -L https://utexas.box.com/shared/static/rhpa8nkcfzpvdguo97n2d5dbn4qb03z8.tar -O models/CONTRIQUE_checkpoint25.tar -q --show-progress
wget -L https://utexas.box.com/shared/static/7s8348b0imqe27qkgq8lojfc2od1631a.tar -O models/CONVIQT_checkpoint10.tar -q --show-progress

Alternatively, the checkpoints can also be downloaded using these links link1 and link2

Google drive link for the checkpoints link1 and link2

Obtaining Quality Scores

We provide trained regressor models in models directory which can be used for predicting image quality using features obtained from CONVIQT model. For demonstration purposes, some sample videos provided in the sample_videos folder.

For blind quality prediction, the following commands can be used.

python3 demo_score.py --video_path sample_videos/30.mp4 --spatial_model_path models/CONTRIQUE_checkpoint25.tar --temporal_model_path models/CONVIQT_checkpoint10.tar --linear_regressor_path models/YouTube_UGC.save
python3 demo_score.py --video_path sample_videos/Flips_crf_48_30fps.webm --spatial_model_path models/CONTRIQUE_checkpoint25.tar --temporal_model_path models/CONVIQT_checkpoint10.tar --linear_regressor_path models/LIVE_YT_HFR.save

Obtaining CONVIQT Features

For calculating CONVIQT features, the following commands can be used. The features are saved in '.npy' format.

python3 demo_feat.py --video_path sample_videos/30.mp4 --spatial_model_path models/CONTRIQUE_checkpoint25.tar --temporal_model_path models/CONVIQT_checkpoint10.tar --features_save_path features.npy
python3 demo_feat.py --video_path sample_videos/Flips_crf_48_30fps.webm --spatial_model_path models/CONTRIQUE_checkpoint25.tar --temporal_model_path models/CONVIQT_checkpoint10.tar --features_save_path features.npy

Training CONVIQT

Download Training Data

Create a directory mkdir training_data to store videos used for training CONVIQT. Run the following commands to download and unzip training data containing videos with synthetic distortions.

bash data_download.sh
bash data_unzip.sh

For UGC videos download Kinetics dataset link and unzip the data. For training CONVIQT only directories parts_0-parts_10 present in Kinetics-400 dataset are needed.

Training Model

Download csv files containing path to videos and corresponding distortion classes.

wget -L https://utexas.box.com/shared/static/63pvroz3287j1kj7ja0gv81vw2txnwlw.csv -O csv_files/file_names_ugc.csv -q --show-progress
wget -L https://utexas.box.com/shared/static/migniec2yb07vc8kz002kub658840dun.csv -O csv_files/file_names_syn.csv -q --show-progress

The above files can also be downloaded manually using these links link1, link2 Google drive links link1, link2

Spatial Feature Extraction

Spatial features are extracted using CONTRIQUE model using the following commands

python3 spatial_feature_extract_syn.py
python3 spatial_feature_extract_ugc.py

Extracted spatial features are saved in training_data directory.

For training CONVIQT with a single GPU the following command can be used

python3 train.py --batch_size 256 --lr 0.6 --epochs 25

Training Linear Regressor

After CONVIQT model training is complete, a linear regressor is trained using CONVIQT features and corresponding ground truth quality scores using the following command.

python3 train_regressor.py --feat_path feat.npy --ground_truth_path scores.npy --alpha 0.1

Contact

Please contact Pavan (pavan.madhusudana@gmail.com) if you have any questions, suggestions or corrections to the above implementation.

Acknowledgement

This repository is built on the CONTRIQUE repository.