Skip to content
/ MDTVSFA Public

[official] Unified Quality Assessment of In-the-Wild Videos with Mixed Datasets Training (IJCV 2021)

License

Notifications You must be signed in to change notification settings

lidq92/MDTVSFA

Repository files navigation

Unified Quality Assessment of In-the-Wild Videos with Mixed Datasets Training

License

Description

MDTVSFA code for the following paper:

How to?

Install Requirements

conda create -n reproducibleresearch pip python=3.6
source activate reproducibleresearch
pip install -r requirements.txt -i https://pypi.tuna.tsinghua.edu.cn/simple
# source deactive

Note: Make sure that the CUDA version is consistent. If you have any installation problems, please find the details of error information in *.log file.

Download Datasets

Download the KoNViD-1k, CVD2014 (alternative link), LIVE-Qualcomm, and LIVE-VQC datasets. Then, run the following ln commands in the root of this project.

ln -s KoNViD-1k_path KoNViD-1k # KoNViD-1k_path is your path to the KoNViD-1k dataset
ln -s CVD2014_path CVD2014 # CVD2014_path is your path to the CVD2014 dataset
ln -s LIVE-Qualcomm_path LIVE-Qualcomm # LIVE-Qualcomm_path is your path to the LIVE-Qualcomm dataset
ln -s LIVE-VQC_path LIVE-VQC # LIVE-VQC_path is your path to the LIVE-VQC dataset

Training and Evaluating on Multiple Datasets

# Feature extraction
CUDA_VISIBLE_DEVICES=0 python CNNfeatures.py --database=KoNViD-1k --frame_batch_size=64
CUDA_VISIBLE_DEVICES=1 python CNNfeatures.py --database=CVD2014 --frame_batch_size=32
CUDA_VISIBLE_DEVICES=0 python CNNfeatures.py --database=LIVE-Qualcomm --frame_batch_size=8
CUDA_VISIBLE_DEVICES=1 python CNNfeatures.py --database=LIVE-VQC --frame_batch_size=8
# Training, intra-dataset evaluation, for example 
chmod 777 job.sh
./job.sh -g 0 -d K -d C -d L > KCL-mixed-exp-0-10-1e-4-32-40.log 2>&1 &
# Cross-dataset evaluation (after training), for example
chmod 777 cross_job.sh
./cross_job.sh -g 1 -d K -d C -d L -c N -l mixed > KCLtrained-crossN-mixed-exp-0-10.log 2>&1 &

Test Demo

The model weights provided in models/MDTVSFA.pt are the saved weights when running the 9-th split of KoNViD-1k, CVD2014, and LIVE-Qualcomm.

python test_demo.py --model_path=models/MDTVSFA.pt --video_path=data/test.mp4

Contact

Dingquan Li, dingquanli AT pku DOT edu DOT cn.

About

[official] Unified Quality Assessment of In-the-Wild Videos with Mixed Datasets Training (IJCV 2021)

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published