The Oxford-IIIT Pet Dataset 转为 YOLO 格式脚本

本脚本用于划分数据集，并由 Oxford-IIIT Pet Dataset 的 Trimap 生成 检测框，其中手动删除了 多个对象的图像，以及 找不到 检测框的图像。

import os
import os.path as osp
from sklearn.model_selection import train_test_split
import pandas as pd
import shutil
import numpy as np
import cv2
 
try:
	import xml.etree.cElementTree as ET
except ImportError:
	import xml.etree.ElementTree as ET
 
def xyxy2xywh(x, shape):
	# Convert boxes from [x1, y1, x2, y2] to [x, y, w, h] where xy1=top-left, xy2=bottom-right
	y = list(np.copy(x))
	y[0] = (x[0] + x[2]) / 2  # x center
	y[1] = (x[1] + x[3]) / 2  # y center
	y[2] = x[2] - x[0]  # width
	y[3] = x[3] - x[1]  # height
 
	# 归一化
	y[0] = round(y[0] / shape[1], 4)
	y[1] = round(y[1] / shape[0], 4)
	y[2] = round(y[2] / shape[1], 4)
	y[3] = round(y[3] / shape[0], 4)
 
	return y
 
def GetAnnotBoxLoc(AnotPath):
	tree = ET.ElementTree(file=AnotPath)  
	root = tree.getroot()
	ObjectSet=root.findall('object')
	ObjBndBoxSet={} 
	bbox_list = []
	for Object in ObjectSet:
		ObjName=Object.find('name').text
		BndBox=Object.find('bndbox')
		x1 = int(BndBox.find('xmin').text)
		y1 = int(BndBox.find('ymin').text)
		x2 = int(BndBox.find('xmax').text)
		y2 = int(BndBox.find('ymax').text)
		bbox_list.append([x1,y1, x2,y2])
	assert len(bbox_list)==1, 'bbox 数目不为1'
	return bbox_list
 
 
 
def deal_data(data, mode):
	cat_number = 0
	dog_number = 0
	for img_name in np.array(data)[:,:3]:
 
		trimaps_img_path = osp.join('annotations/trimaps', img_name[0] + '.png')
		AnotPath = osp.join('annotations/xmls', img_name[0] + '.xml')
		in_img_path = osp.join('images', img_name[0] + '.jpg')
		if os.path.exists(trimaps_img_path) and os.path.exists(AnotPath) and os.path.exists(in_img_path):
			print(in_img_path)
 
			if img_name[2]==1: # cat
				cat_number += 1
				# cls = str(3)
				cls = 'cat'
			else: # dog
				dog_number += 1 
				# cls = str(4)
				cls = 'dog'
 
			trimaps_img = cv2.imread(trimaps_img_path, 0)
			n = 1
			xy = np.where(trimaps_img == n)
			y = xy[0]
			x = xy[1]
 
			x1,y1, x2,y2 =  x.min(), y.min(), x.max(), y.max() # cat/dog bbox 左上，右下
 
			# original_img = cv2.imread(in_img_path)
			# temp_image = cv2.rectangle(original_img, (x1, y1), (x2, y2), (0, 0, 255), 2) # （左上，右下）
 
			shape = trimaps_img.shape # h, w
			body_bbox = xyxy2xywh((x1,y1, x2,y2), shape)
			bbox_list = GetAnnotBoxLoc(AnotPath)
			with open(osp.join('../Pet/labels/' + mode, 'Oxford_IIIT_Pet_' + img_name[0] + '.txt'),mode = "w", encoding = "utf-8") as f:
				f.write(cls + ' ')
				for head_bbox in bbox_list:
					# temp_image = cv2.rectangle(temp_image, (head_bbox[0], head_bbox[1]), (head_bbox[2], head_bbox[3]), (0, 0, 255), 2)
					head_bbox = xyxy2xywh(head_bbox, shape)
					for bbx in body_bbox + head_bbox:
						f.write(str(bbx) + ' ')
					f.write('\n')   
 
			# plot_img_path = osp.join('plot_body_images/', img_name[0] + '.jpg')
			# cv2.imwrite(plot_img_path, temp_image)
 
			out_img_path = osp.join('../Pet/images/' + mode, 'Oxford_IIIT_Pet_' + img_name[0] + '.jpg')
			shutil.copy(in_img_path, out_img_path)
 
	return cat_number, dog_number
def break_data(origin_dataset, rate=0.2):
	train_data, test_data = train_test_split(origin_dataset, test_size=rate)
	train_data, eval_data = train_test_split(train_data, test_size=rate)
 
	train_cat_number, train_dog_number = deal_data(train_data, mode='train')
	val_cat_number, val_dog_number = deal_data(eval_data, mode='val')
	test_cat_number, test_dog_number = deal_data(test_data, mode='test')
 
	print('train_data:',len(train_data))
	print('test_data:',len(test_data))
	print('eval_data:',len(eval_data))
 
	print('cat_number:', train_cat_number + val_cat_number + test_cat_number, 'dog_number:', train_dog_number + val_dog_number + test_dog_number)
	print(train_cat_number + val_cat_number + test_cat_number+train_dog_number + val_dog_number + test_dog_number)
if __name__ == "__main__":
 
	data = pd.read_csv('annotations/list.txt', header=None, sep=' ')
	break_data(origin_dataset=data, rate=0.1)