Quality Scoring

The quality scoring for each dataset is crucial to understanding the quality of the dataset. On this page you can find out how the metric scores are calculated. As Valyu Exchange continues to mature, we aim to add more metrics with indivdual scores instead of a general score.

Text Quality

The text quality score is computed using several classification models. Text quality metrics such as accurate spelling and grammar affect the score. The score is weighted using similarity against written texts to ensure formality. Accuracy of factual information within the text also leads to a higher scoring.

Image Quality

The image quality score is computed using a classification model trained on several labeled image datasets. The model scores an image on objective qualities such as blurriness, noise, artefacts, resolution, watermarks and entropy.

Video Quality

The image quality score is computed using a classification model trained on several labeled video datasets. The model scores an image on objective qualities such as blurriness, noise, artefacts, watermarks, entropy, watermarks and frame rate.

Audio Quality

The audio quality score is computed using a classification model trained on several labeled audio datasets. The model evaluates audio samples based on objective qualities such as clarity, noise levels, bitrate, sample rate, and the presence of artifacts or distortions.