YOLOv7数据预处理部分代码（bbox归一化及数据路径生成）.zip

import os import cv2 ''' 将注释文件转换为yolo格式的训练文件 ''' label2int = {'banana': 0, 'CARDBOARD': 1, 'bottle': 2} def convert(c, size, box): dw, dh = 1. / (size[0]), 1. / (size[1]) x = round((box[0] + box[2]) * 0.5 * dw, 8) y = round((box[1] + box[3]) * 0.5 * dh, 8) w = round((box[2] - box[0]) * dw, 8) h = round((box[3] - box[1]) * dw, 8) return [c, x, y, w, h] def annosNoormalize(train_image_label: str, train_image_dir: str, train_annos: str): for idx, img in enumerate(os.listdir(train_image_label)): # 得到图片和注释文件路径 img_path = train_image_dir + img[:-4] + '.jpg' label_path = train_image_label + img # 读取图片得到图片尺寸信息:416*416 or 640*640 image = cv2.imread(img_path) size = image.shape[:2] # width,height # 读取注释文件得到bbox信息 annos = [] with open(label_path, 'rb') as f: # 文件读取内容为bytes类型字符串，需要用decode函数解码 for line in f.readlines(): line = line.decode().strip().split(' ') c = label2int[line[0]] # 字符型标签转为整形标签 bbox = list(map(float, line[4:8])) # x_min, y_min, x_max, y_max anno = convert(c, size, bbox) annos.append(anno) # 将bbox信息写入到新注释文件 with open(train_annos + img, 'a+') as f: for items in annos: for num, item in enumerate(items): f.write(str(item) if num == 0 else ' ' + str(item)) f.write('\n') print("当前已经处理完成{}的{}个注释文件".format(train_annos.split('_')[0], idx + 1)) if __name__ == '__main__': train_image_dir = 'data/train/images/' train_image_label = 'data/train/labels/' train_annos = 'train_labels/' val_image_dir = 'data/val/images/' val_image_label = 'data/val/labels/' val_annos = 'val_labels/' annosNoormalize(train_image_label, train_image_dir, train_annos) annosNoormalize(val_image_label, val_image_dir, val_annos)