Extraction Configuration

To extract features from data for use with Cineast, an extraction job config JSON file must be used. A few simple example extraction job configuration files can be found here.

Here is a very basic example of such a job configuration file:

{
	"input": {
		"path": "/path/to/data",
		"depth": 2,
		"id": {
			"name": "SequentialObjectIdGenerator",
			"properties": {}
		}
	},
	"extractors": [
		{"name": "AverageColor"},
		{"name": "MedianColor"}
	],
	"exporters": [
		{
			"name": "ShotThumbnailsExporter",
			"properties": {
				"destination":"/output/path/for/thumbnails"
			}
		}
	],
	"database": {
		"writer": "COTTONTAIL",
		"selector": "COTTONTAIL"
	}
}

The extraction config file is made up of four main parts: input, extractors, exporters and database.

Input

The most important parts of the input object are:

• path: The path to the source data directory.
• depth: The depth up to which the data directory should be explored for data (e.g. a depth value of 2 searches the files in the specified directory and immediate subdirectories).
• id: An object containing the classname of the desired object ID generator and related properties. The available ID generators can be found in the idgenerator package.

Extractors

The extractors list contains objects each containing the name of feature extractor class. The available feature extractors can be found in the features package.

Exporters

The exporters list contains objects each with the name of an exporter class and associated properties. The available exporters can be found in the exporter package. These are used to export additional data, such as thumbnails in the example above.

Database

The database object contains information on where the extracted data should be written, by specifying a writer and a selector option. The writer is responsible for writing the extracted data, while the selector is responsible for making sure that data that already exists in the output location (in case of e.g. a database) is not written a second time. The available options can be found in the DatabaseConfig class.

The above example would attempt to write the extracted data directly into a Cottontail DB instance. To instead write extracted data as JSON files, the database object would have to be replaced with:

"database":{
	"selector": "NONE",
	"writer": "JSON",
	"host": "/output/path/for/extracted/data/jsons"
}

Type

By default, Cineast tries to extract all types of media it finds in the specified data location and will apply each feature extractor to all file types it supports. If you wish to only extract a specific type of media, e.g. image sequences, then specify the type in the top-level config, e.g.:

"type": "IMAGE_SEQUENCE"

For a full list of supported types, see org.vitrivr.cineast.core.data.MediaType.

Image Sequences

Image sequences are an exception to the default extraction behavior in that they are not extracted by default and must be explicitly specified as type. In the absence of the type field, image sequence directories are treated as directories of independent images. Since directories of independent images may be indistinguishable from image sequence directories, these two types of directories must be extracted through separate extraction configurations to avoid one type being processed as the other.