”dis" and "ref" in dataset json file about dvqa HOT 3 CLOSED

Hi @ymxcywl ,

1. The ref and dis must have same size as input. The ref, an abbreviation of reference, is the lossless version of a vide. Then dis, an abbreviation of distorted, is corresponding distorted version of aforementioned ref.

2. Originally we have lossless ref.yuv, then we compress it to dis.mp4, and one would want to predict the MOS of dis.mp4.

3. If you just want to evaluate on your own video, simply create dataset.json with the following format. Probably, you need to conduct subjective test to get the MOS, i.e. ground truth, of your videos.

{
    "test": {
        "dis": ["dis_1.yuv", "dis_2.yuv"],
        "ref": ["ref_1.yuv", "ref_2.yuv"],
        "fps": [30, 24],
        "mos": [94.2, 55.8],
        "height": [1080, 720],
        "width": [1920, 1280]
    },
    "train": {
        "dis": ["dis_3.yuv", "dis_4.yuv"],
        "ref": ["ref_3.yuv", "ref_4.yuv"],
        "fps": [50, 24],
        "mos": [85.2, 51.8],
        "height": [320, 720],
        "width": [640, 1280]
    }
}

from dvqa.

Thank you so much.
So, is DVQA a full reference assessment?
If it is difficult to get orignal lossless version of a video, is it reasonable to use DVQA to predict mos ?(eg: dis is the video of low bps, while ref is the original mp4 video)

from dvqa.

Yes. DVQA is a FR metric.

If you have the original.mp4 video in hand, and suppose it was losslessly compressed from origihal.yuv, all you need to do is decode it back to yuv format. A decoding script is provided at decode_stream.py

from dvqa.