core.services.model.evaluator.utils.coco_utils

coco_utils

Functions:

Name	Description
load_json	Load and return the contents of a JSON file.
save_json	Save data to a JSON file with an indentation of 4 spaces.
adjust_image_ids	Adjust image IDs in COCO data. If the image IDs start at 0, they are incremented by 1.
renumber_annotation_ids	Renumber annotation IDs sequentially starting from 1.
fix_coco_ids	Fix image and annotation IDs in a COCO file. Images whose IDs start at 0 are adjusted and
create_image_id_mapping	Create a mapping between image IDs from the ground truth and prediction data based on the 'file_name' field.
fix_image_ids	Fix the 'image_id' fields in prediction data using the provided mapping.
match_image_ids	Match image IDs between the ground truth and prediction files.
compute_tp_fp_fn	Compute the number of True Positives (TP), False Positives (FP), and False Negatives (FN) per category
calculate_metrics	Calculate precision, recall, and F1-score.
evaluate_category	Evaluate predictions for a given category using COCO metrics.

load_json(file_path)

Load and return the contents of a JSON file.

Parameters:

Name	Type	Description	Default
file_path	str	Path to the JSON file.	required

Returns:

Name	Type	Description
dict	dict	Parsed JSON data.

save_json(data, file_path)

Save data to a JSON file with an indentation of 4 spaces.

Parameters:

Name	Type	Description	Default
data	dict	Data to save.	required
file_path	str	Output path for the JSON file.	required

adjust_image_ids(coco_data)

Adjust image IDs in COCO data. If the image IDs start at 0, they are incremented by 1.

Parameters:

Name	Type	Description	Default
coco_data	dict	COCO data containing the "images" and "annotations" keys.	required

renumber_annotation_ids(coco_data)

Renumber annotation IDs sequentially starting from 1.

Parameters:

Name	Type	Description	Default
coco_data	dict	COCO data containing the "annotations" key.	required

fix_coco_ids(coco_path)

Fix image and annotation IDs in a COCO file. Images whose IDs start at 0 are adjusted and annotation IDs are renumbered sequentially. The fixed file is saved with the suffix '_fixed'.

Parameters:

Name	Type	Description	Default
coco_path	str	Path to the original COCO file.	required

Returns:

Name	Type	Description
str	str	Path to the fixed COCO file.

create_image_id_mapping(gt_images, pred_images)

Create a mapping between image IDs from the ground truth and prediction data based on the 'file_name' field.

Parameters:

Name	Type	Description	Default
gt_images	list	List of ground truth images (each a dict with 'file_name' and 'id').	required
pred_images	list	List of predicted images (each a dict with 'file_name' and 'id').	required

Returns:

Name	Type	Description
dict	dict	Mapping {predicted_image_id: ground_truth_image_id}.

fix_image_ids(pred_data, id_mapping)

Fix the 'image_id' fields in prediction data using the provided mapping.

Parameters:

Name	Type	Description	Default
pred_data	dict	Prediction COCO data containing "images" and "annotations".	required
id_mapping	dict	Mapping between predicted and ground truth image IDs.	required

match_image_ids(ground_truth_file, prediction_file, corrected_prediction_file)

Match image IDs between the ground truth and prediction files. Loads the ground truth and prediction JSON files, creates a mapping based on 'file_name', fixes the prediction image IDs using the mapping, and saves the corrected predictions.

Parameters:

Name	Type	Description	Default
ground_truth_file	str	Path to the COCO ground truth file.	required
prediction_file	str	Path to the COCO predictions file.	required
corrected_prediction_file	str	Output path for the corrected predictions file.	required

compute_tp_fp_fn(coco_eval)

Compute the number of True Positives (TP), False Positives (FP), and False Negatives (FN) per category using a COCOeval object.

Parameters:

Name	Type	Description	Default
coco_eval	COCOeval	COCOeval object after evaluation.	required

Returns:

Name	Type	Description
dict	dict	Dictionary mapping each category ID to a dict with keys "TP", "FP", and "FN".

calculate_metrics(tp, fp, fn)

Calculate precision, recall, and F1-score.

Parameters:

Name	Type	Description	Default
tp	int	Number of True Positives.	required
fp	int	Number of False Positives.	required
fn	int	Number of False Negatives.	required

Returns:

Name	Type	Description
tuple	tuple[float, float, float]	Precision, recall, and F1-score.

evaluate_category(coco_gt, coco_pred, cat_name, inference_type)

Evaluate predictions for a given category using COCO metrics. Runs evaluation for the specified category and area 'all', then computes additional metrics (TP, FP, FN, precision, recall, F1-score).

Parameters:

Name	Type	Description	Default
coco_gt	COCO	COCO object for the ground truth.	required
coco_pred	COCO	COCO object for the predictions.	required
cat_name	str	Category name.	required
inference_type	InferenceType	Type of inference (classification, detection, or segmentation).	required

Returns:

Name	Type	Description
dict	dict	Dictionary containing evaluation metrics for the category.