VEDIA数据集处理

先分享三个数据集的网址，第一个是我下数据集的，后两个是用来改数据集的标签格式的github，一个是darknet格式，一个是imagenet格式

vedia数据集下载地址

github

github2

我用的代码是yolov3，所以我需要darknet格式的标签

就是类别，加四个坐标参数，我修改了github2的代码，我只需要其中七类，所以我把其他类别的标签删了

 

# 1: car, 2:trucks, 4: tractors, 5: camping cars, 7: motorcycles, 8:buses, 9: vans, 10: others, 11: pickup, 23: boats , 201: Small Land Vehicles, 31: Large land Vehicles

import os
import pandas as pd


def update_annotations(filename):
	# with open('E:/数据集/VEDAI/annotation_ori/' + filename, 'r') as file:
	# 	data = file.read()

		# text = '\n'.join(' '.join(line.split()) for line in data.split('\n'))
		# text.replace('\t', ' ')

		# print(text)
		# output = open('E:/数据集/VEDAI/label/' + filename, 'w')
		# output.write(text)
		# output.close()

	data = pd.read_csv('E:/数据集/VEDAI/annotation_ori/' + filename, sep=' ', index_col=None, header=None, names=['x_center', 'y_center', 'orientation', 'class', 'is_contained', 'is_occluded', 'corner1_x', 'corner2_x', 'corner3_x', 'corner4_x', 'corner1_y', 'corner2_y', 'corner3_y', 'corner4_y'])

	# data['class'].replace(11, 3, inplace=True)
	# data['class'].replace(23, 6, inplace=True)
	# data['class'].replace(201, 11, inplace=True)
	# data['class'].replace(31, 12, inplace=True)
	data['class'].replace(1, 0, inplace=True)
	data['class'].replace(11, 1, inplace=True)
	data['class'].replace(2, 3, inplace=True)
	data['class'].replace(5, 2, inplace=True)
	data['class'].replace(4, 5, inplace=True)
	data['class'].replace(10, 4, inplace=True)
	data['class'].replace(23, 6, inplace=True)
	data['class'].replace(9, 7, inplace=True)
	# data['class'].replace(31, 9, inplace=True)
	# data['class'].replace(8, 8, inplace=True)
	# print(data['class'])

	# data['class'] = data['class'] - 1
	# data['x_center_ratio'] = data['x_center'].astype(float) / 1024.
	# data['y_center_ratio'] = data['y_center'].astype(float) / 1024.
	data['x_center_ratio'] = data['x_center'].astype(float) / 512.
	data['y_center_ratio'] = data['y_center'].astype(float) / 512.
	# data['width_ratio'] = (data[['corner1_x', 'corner2_x', 'corner3_x', 'corner4_x']].max(axis=1) - data[['corner1_x', 'corner2_x', 'corner3_x', 'corner4_x']].min(axis=1)) / 1024.
	# data['height_ratio'] = (data[['corner1_y', 'corner2_y', 'corner3_y', 'corner4_y']].max(axis=1) - data[['corner1_y', 'corner2_y', 'corner3_y', 'corner4_y']].min(axis=1)) / 1024.
	data['width_ratio'] = (data[['corner1_x', 'corner2_x', 'corner3_x', 'corner4_x']].max(axis=1) - data[['corner1_x', 'corner2_x', 'corner3_x', 'corner4_x']].min(axis=1)) / 512.
	data['height_ratio'] = (data[['corner1_y', 'corner2_y', 'corner3_y', 'corner4_y']].max(axis=1) - data[['corner1_y', 'corner2_y', 'corner3_y', 'corner4_y']].min(axis=1)) / 512.

	res = data.drop(['x_center', 'y_center', 'corner1_x', 'corner2_x', 'corner3_x', 'corner4_x', 'orientation', 'corner1_y', 'corner2_y', 'corner3_y', 'corner4_y', 'is_contained', 'is_occluded'], axis=1)
	res = res.drop(index=res.loc[(res['class'] >7)].index) #删除不要的标签
	# print(res)
	# if data['class'][0]>7:
	# 	print(data['class'][0],filename)
	# for i in range(len(data['class'])):
	# 	if data['class'][i]>7:
	# 		print(data['class'][i],filename)
	# if filename == '00000043.txt':
	# 	print(data['class'][0])
	# 	output = open('E:/数据集/VEDAI/labels/' + filename, 'w')
	# 	output.close()
	# else:
	res.to_csv('E:/数据集/VEDAI/labels/' + filename, sep=' ', index=False, header=None)


list = os.listdir('E:/数据集/VEDAI/annotation_ori/')
# list.remove('.DS_Store')  # for macOS
for filename in list:
	# print(filename)
	update_annotations(filename)