MFQEv2 Dataset

For some video enhancement/restoration tasks, lossless reference videos are necessary.

We open-source the dataset used in our MFQEv2 paper, which includes 108 lossless YUV videos for training and 18 test videos recommended by ITU-T.

1. Content

• 108 lossless YUV videos for training.
• 18 lossless YUV videos for test, recommended by ITU-T.
• An HEVC compression tool box.

43.1 GB in total.

2. Download Raw Videos

[Dropbox]

or [百度网盘 (key: mfqe)]

3. Compress Videos

We compress both training and test videos by HM 16.5 at low delay P (LDP) mode with QP=37. The video compression toolbox is provided at the dataset folder.

We will get:

MFQEv2_dataset/
├── train_108/
│   ├── raw/
│   └── HM16.5_LDP/
│       └── QP37/
├── test_18/
│   ├── raw/
│   └── HM16.5_LDP/
│       └── QP37/
├── video_compression/
│   └── ...
└── README.md

Ubuntu

1. cd video_compression/
2. Edit option.yml.
3. chmod +x TAppEncoderStatic
4. python unzip_n_compress.py

Windows

1. Unzip train_108.zip and test_18.zip manually!
2. cd video_compression\
3. Edit option.yml (e.g., system: windows).
4. python unzip_n_compress.py

If you find this dataset helpful, please cite our paper.