idtrackerai

Human validation

It is possible to perform Human validation in two different ways. On one hand, we intend with global a human generated ground truth in which the identities of all the animals in a frame are validated. This kind of validation is extremely hard when dealing with large groups. To verify the goodness of the tracking on the entire video the individual validation allows to follow a single selected animals.

Generate groundtruth

class generate_groundtruth.GroundTruthBlob(attributes_to_get=['identity', 'assigned_identity', 'used_for_training', 'accumulation_step', 'centroid', 'pixels', 'frame_number', 'is_an_individual', 'is_a_crossing', 'was_a_crossing', 'blob_index', 'fragment_identifier'])[source]

Lighter blob objects. Attributes:

identity (preferring the one assigned by the user, if it is not None) centroid pixels (pixels is stored to check the groundtruth in crossings)

Methods

get_attribute(blob)
generate_groundtruth.generate_groundtruth(video, blobs_in_video=None, start=None, end=None, wrong_crossing_counter=None, unidentified_individuals_counter=None, save_gt=True)[source]

Generates a list of light blobs_in_video, given a video object corresponding to a tracked video

Compute ground truth statistics

compute_groundtruth_statistics.compare_tracking_against_groundtruth_no_gaps(number_of_animals, groundtruth, blobs_in_video_groundtruth, blobs_in_video, identities_dictionary_permutation)[source]

blobs_in_video_groundtruth: cut groundtruth.blobs_in_video using start and end of the groundtruth object blobs_in_video: cut list_of_blobs.blobs_in_video using start and end of the groundtruth object

Generate individual ground truth

class generate_individual_groundtruth.GroundTruthBlob(attributes_to_get=['identity', 'assigned_identity', 'used_for_training', 'accumulation_step', 'centroid', 'pixels', 'frame_number', 'is_an_individual', 'is_a_crossing', 'blob_index', 'fragment_identifier'])[source]

Lighter blob objects. Attributes:

identity (preferring the one assigned by the user, if it is not None) centroid pixels (pixels is stored to check the groundtruth in crossings)

Methods

get_attribute(blob)
generate_individual_groundtruth.generate_individual_groundtruth(video, blobs_in_video=None, start=None, end=None, validated_identity=None, save_gt=True)[source]

Generates a list of light blobs_in_video, given a video object corresponding to a tracked video

Compute individual ground truth statistics