Viewing Dataset Statistics
To view dataset metrics, select one or more datasets and click one of the available dataset analytics icons. You can toggle between different metrics and views, and analyze different combinations of datasets by hitting the refresh icon after selecting or deselecting a dataset. Use the available icons to print, view in fullscreen, or download a screenshot or .csv containing the raw data.
These same analytics are available from the dataset details page to view these metrics for a subset a particular dataset.
Color metrics are calculated using the pixel values for each channel - typically red, green, and blue - or all three, in the case of grey values. The various graphs - scatter plots, histograms, and candlestick plots - show the color distribution of the selected dataset(s) in different ways, in order to highlight outliers and common values. Click an individual data point or bar to view the specific images or videos with those values.
Innotescus provides a quality metric, BRISQUE - Blind/Referenceless Image Spatial Quality Evaluator, to view the quality distribution of items in a dataset. Users can view the distribution of this metric as a candlestick plot or bar chart, and explore the items in each bin by clicking on a bar in the histogram.
Innotescus provides a complexity metric, entropy, to show the distribution of complexity within a dataset. The entropy calculation encompasses things like varying textures or noise within an image or frame of a video, and is used to inform decisions about the complexity of the network architecture being used. View this distribution as a histogram or candlestick plot, and click on the bars of the histogram to view the items in a certain bin.
Use dive to create a custom view of your dataset items with the metrics available above. Use the binning and scatter fields to create different views, or combine them to visualize across four dimensions of data. If there are few enough data points, each point will show as a thumbnail of the image it represents.
Access task-specific metadata in your dive visualization by accessing the dive analytics within a task's datasets tab.
Dive contains a few metrics that are not included in the color, quality, or complexity charts - brightness, lightness, hue, and saturation. To learn more about these standard metrics, see this article.