extract.FeatureExtractor

extract.FeatureExtractor(batch_size, device=None, stochastic=False)

Extract features from images.

Extract feature vectors for individual identification from images. Currently, FeatureExtractor only includes the AnyDorsal algorithm.

Parameters

Name	Type	Description	Default
batch_size	int	The number of images the GPU will process.	required
device	(None, cuda, mps, cpu)	Device with which to extract the features. By default, the best device is chosen for the user (cuda, mps, or cpu)	`None`
stochastic	boolean	Currently unused.	`False`

Examples

For a complete working example with real images, see:

Tutorial

Basic usage pattern::

from pyseter.extract import FeatureExtractor

# Initialize extractor
extractor = FeatureExtractor(batch_size=16)

# Extract features from all images
features = extractor.extract('path/to/images/')

# Access individual image features
img_features = features['my_image.jpg']

Methods

Name	Description
extract	Extracts features from images.

extract

extract.FeatureExtractor.extract(image_dir, bbox_csv=None)

Extracts features from images.

Extracts feature vectors for every image in a directory with the AnyDorsal algorithm.

Parameters

Name	Type	Description	Default
image_dir	str	Directory of images to from which to extract features. Directory should be flat, in that there should not be subdirectories with images.	required
bbox_csv	str	Optional path to csv file with bounding boxes for each image in the image_dir.	`None`

Returns

Name	Type	Description
	dict	A mapping image file names to the corresponding feature vector. The file names are represented as strings, while the feature vector. is a NumPy array. For example: `{'img1.jpg': np.array([0.1, 0.1, 0.2, ..., 0.9]), 'img2.jpg': np.array([0.2, 0.3, 0.4, ..., 0.1])}` The numpy array should have length 5504.

Raises

Name	Type	Description
	OutOfMemoryError	The GPU has run out of memory. Try reducing your batch size, or reducing the file size of the images in the directory.