VQML is No Reference AI based Video Quality Measurement Solution. It predicts the quality of the video itself and it measures on a scale of 1~5, corresponding to real-world human perception.
Measurement based on Deep Learning
To solve the time and cost waste, VQML™ does continuous Deep Learning for measurement |
Learning by Highly Reliable Database
VQML™ has learned pattern about videos and MOS based on the database through the viewer survey |
No Reference Solution
Measurement values are 1 to 5 like MOS, VQML™ needs only what a measurement video |
Learning Data
KoNViD 1k + XCAL Dataset |
Learning Type
Supervised learning |
Development/Utilizationn Model
2 CNN Layer + 1 GRU Layer |
Technology Application/API
PyTorch |
|
Technology applied to
XCAL, VQML-Server |
| Version | Correlation | MAE |
|---|---|---|
| GPU | 86.5 | 0.235 |
| No-GPU | 84.1 | 0.259 |
※ Test dataset: KoNViD-1k dataset
