run_evaluation.py

import cv2
import numpy as np
import glob
import os
from pathlib import Path
import json
from preprocessing.preprocess import Preprocess
from metrics.evaluation import Evaluation


class EvaluateAll:

    def __init__(self):
        os.chdir(os.path.dirname(os.path.realpath(__file__)))

        with open('config.json') as config_file:
            config = json.load(config_file)

        self.images_path = config['images_path']
        self.annotations_path = config['annotations_path']

    def get_annotations(self, annot_name):
        with open(annot_name) as f:
            lines = f.readlines()
            annot = []
            for line in lines:
                l_arr = line.split(" ")[1:5]
                l_arr = [int(i) for i in l_arr]
                annot.append(l_arr)
        return annot

    def run_evaluation(self):

        im_list = sorted(glob.glob(self.images_path + '/*.png', recursive=True))
        iou_arr = []
        preprocess = Preprocess()
        evaluation = Evaluation()

        # Change the following detector and/or add your detectors below

        # CASCADE
        import detectors.cascade_detector.detector as cascade_detector
        # cascade_detector = cascade_detector.Detector()

        # V2
        import detectors.v2.detector as v2
        v2 = v2.Detector()

        # V3
        import detectors.v3.detector as v3
        v3 = v3.Detector()

        # V4
        import detectors.v4.detector as v4
        v4 = v4.Detector()

        chosen_detector = v4

        for im_name in im_list:
            # Read an image
            img = cv2.imread(im_name)

            # Apply some preprocessing
            # img = preprocess.histogram_equlization_rgb(img)  # This one makes VJ worse

            # Run the detector. It runs a list of all the detected bounding-boxes. In segmentor you only get a mask matrices, but use the iou_compute in the same way.
            prediction_list = chosen_detector.detect(im_name)

            # Read annotations:
            annot_name = os.path.join(self.annotations_path, Path(os.path.basename(im_name)).stem) + '.txt'
            annot_list = self.get_annotations(annot_name)

            # Only for detection:
            p, gt = evaluation.prepare_for_detection(prediction_list, annot_list)

            iou = evaluation.iou_compute(p, gt)
            iou_arr.append(iou)

        miou = np.average(iou_arr)
        print("\n")
        print("Average IOU:", f"{miou:.2%}")
        print("\n")


if __name__ == '__main__':
    ev = EvaluateAll()
    ev.run_evaluation()