test_detection.py

from pathlib import Path

import numpy as np
import pytest
import debobo


@pytest.mark.parametrize('idx, iou_thresh, max_detections', pytest.coco_params)
def test_map_score_evaluator(idx, iou_thresh, max_detections, detection_data, coco_results):
    image_ids = list(set(detection_data.dt['image_id']).union(detection_data.gt['image_id']))
    groundtruths = (
        detection_data.gt.loc[detection_data.gt['image_id'] == image_id] \
        .to_records(index=False)[['x1', 'y1', 'x2', 'y2', 'class']]
        for image_id in image_ids)
    detections = (
        detection_data.dt.loc[detection_data.dt['image_id'] == image_id]
        .to_records(index=False)[['x1', 'y1', 'x2', 'y2', 'class', 'score']]
        for image_id in image_ids)

    rank_arrays = debobo.evaluate_frames(
        groundtruths, detections, iou_thresh=iou_thresh,
        max_detections=max_detections)

    # make sure mAP score agrees with
    y = debobo.mean_average_precision_score(rank_arrays)
    np.testing.assert_almost_equal(y, coco_results[idx], decimal=2)

    # make sure all classes have a number
    classes = set(detection_data.gt['class']).union(detection_data.dt['class'])
    assert set(rank_arrays) == classes