wider_extract.py

import numpy as np


def make_npy(annotation_file, relative_img_path):
    face_bb = {}
    with open(annotation_file) as f:
        l_filename = True
        l_bb_cnt = False
        l_bb = 0
        f_name = " "
        bb = []

        bb_cnt = -1

        for line in f:
            if bb_cnt == 0:
                l_bb = False
                l_filename = True
                bb_cnt = -1

            if l_filename:
                f_name = line[:-1]
                l_bb_cnt = True
                l_filename = False
                # print ("f_name : ", f_name)
                continue

            if l_bb_cnt:
                bb_cnt = int(line)
                bb_ref_cnt = bb_cnt
                l_bb = True
                l_bb_cnt = False
                # print ("bb_cnt : ", bb_cnt)
                continue

            if l_bb:
                if bb_ref_cnt == bb_cnt:
                    bb = [relative_img_path + f_name, relative_img_path + f_name, [300, 300]]
                    # bb.append(bb_ref_cnt)

                if bb_cnt > 0:
                    bb_cnt -= 1
                    bb_each = []
                    bb_info = line.split(" ")  # x1, y1, w, h, blur, expression, illumination, invalid, occlusion, pose

                    isValid = bb_info[7]  # 0, 1 using only 0
                    isOccluded = bb_info[8]  # 0,1,2 use only 0,1
                    isBlur = bb_info[4]  # 0,1,2 use only 0,1

                    x1 = int(bb_info[0])
                    x2 = x1 + int(bb_info[2])
                    y1 = int(bb_info[1])
                    y2 = y1 + int(bb_info[3])

                    bb_each.append([x1, x2, y1, y2])
                    class_id = 1  # face
                    bb_each.append(class_id)

                    if isValid == 0 and isOccluded != 2 and isBlur != 2:
                        bb.append(bb_each)
                        # print ("bb: " , bb)
                        # print ("line : ", line)

                if bb_cnt == 0:
                    face_bb[f_name] = bb

        np.save('wider_test.npy', face_bb)


if __name__ == '__main__':
    train_annotation_file = "dataset/wider_face_train_bbx_gt.txt"
    val_annotation_file = "dataset/wider_face_val_bbx_gt.txt"
    train_img_path = "dataset/WIDER_train/images/"
    val_img_path = "dataset/WIDER_val/images/"

    make_npy(train_annotation_file, train_img_path)
    make_npy(val_annotation_file, val_img_path)