Visdrone预处理
本篇博客包含：
- 切割数据集
- 将数据做成Yolo格式
- 图像二值化
- 绘制评估曲线
github：https://github.com/mary-0830/visdrone-dataset
一、把Visdrone数据集进行切割，生成600*600大小的图片和xml文件
train_crop_visdrone.py
import os
import scipy.misc as misc
from xml.dom.minidom import Document
import numpy as np
import copy, cv2
def save_to_txt(save_path, objects_axis):
f = open(save_path,'w')
objects_list = objects_axis.tolist()
objects_ = [','.join(map(str, i)) + '\n' for i in objects_list]
objects_[-1] = objects_[-1][:-1]
# import pdb
# pdb.set_trace()
f.writelines(objects_)
f.close()
def save_to_xml(save_path, im_width, im_height, objects_axis, label_name, name, hbb=True):
im_depth = 0
object_num = len(objects_axis)
doc = Document()
annotation = doc.createElement('annotation')
doc.appendChild(annotation)
folder = doc.createElement('folder')
folder_name = doc.createTextNode('Visdrone')
folder.appendChild(folder_name)
annotation.appendChild(folder)
filename = doc.createElement('filename')
filename_name = doc.createTextNode(name)
filename.appendChild(filename_name)
annotation.appendChild(filename)
source = doc.createElement('source')
annotation.appendChild(source)
database = doc.createElement('database')
database.appendChild(doc.createTextNode('The Visdrone Database'))
source.appendChild(database)
annotation_s = doc.createElement('annotation')
annotation_s.appendChild(doc.createTextNode('Visdrone'))
source.appendChild(annotation_s)
image = doc.createElement('image')
image.appendChild(doc.createTextNode('flickr'))
source.appendChild(image)
flickrid = doc.createElement('flickrid')
flickrid.appendChild(doc.createTextNode('322409915'))
source.appendChild(flickrid)
owner = doc.createElement('owner')
annotation.appendChild(owner)
flickrid_o = doc.createElement('flickrid')
flickrid_o.appendChild(doc.createTextNode('knautia'))
owner.appendChild(flickrid_o)
name_o = doc.createElement('name')
name_o.appendChild(doc.createTextNode('yang'))
owner.appendChild(name_o)
size = doc.createElement('size')
annotation.appendChild(size)
width = doc.createElement('width')
width.appendChild(doc.createTextNode(str(im_width)))
height = doc.createElement('height')
height.appendChild(doc.createTextNode(str(im_height)))
depth = doc.createElement('depth')
depth.appendChild(doc.createTextNode(str(im_depth)))
size.appendChild(width)
size.appendChild(height)
size.appendChild(depth)
segmented = doc.createElement('segmented')
segmented.appendChild(doc.createTextNode('0'))
annotation.appendChild(segmented)
for i in range(object_num):
objects = doc.createElement('object')
annotation.appendChild(objects)
object_name = doc.createElement('name')
object_name.appendChild(doc.createTextNode(label_name[int(objects_axis[i][5])]))
objects.appendChild(object_name)
pose = doc.createElement('pose')
pose.appendChild(doc.createTextNode('Unspecified'))
objects.appendChild(pose)
truncated = doc.createElement('truncated')
truncated.appendChild(doc.createTextNode('1'))
objects.appendChild(truncated)
difficult = doc.createElement('difficult')
difficult.appendChild(doc.createTextNode('0'))
objects.appendChild(difficult)
bndbox = doc.createElement('bndbox')
objects.appendChild(bndbox)
if hbb:
x0 = doc.createElement('xmin')
x0.appendChild(doc.createTextNode(str((objects_axis[i][0]))))
bndbox.appendChild(x0)
y0 = doc.createElement('ymin')
y0.appendChild(doc.createTextNode(str((objects_axis[i][1]))))
bndbox.appendChild(y0)
x1 = doc.createElement('xmax')
x1.appendChild(doc.createTextNode(str((objects_axis[i][2]))))
bndbox.appendChild(x1)
y1 = doc.createElement('ymax')
y1.appendChild(doc.createTextNode(str((objects_axis[i][3]))))
bndbox.appendChild(y1)
else:
x0 = doc.createElement('x0')
x0.appendChild(doc.createTextNode(str((objects_axis[i][0]))))
bndbox.appendChild(x0)
y0 = doc.createElement('y0')
y0.appendChild(doc.createTextNode(str((objects_axis[i][1]))))
bndbox.appendChild(y0)
x1 = doc.createElement('x1')
x1.appendChild(doc.createTextNode(str((objects_axis[i][2]))))
bndbox.appendChild(x1)
y1 = doc.createElement('y1')
y1.appendChild(doc.createTextNode(str((objects_axis[i][3]))))
bndbox.appendChild(y1)
x2 = doc.createElement('x2')
x2.appendChild(doc.createTextNode(str((objects_axis[i][4]))))
bndbox.appendChild(x2)
y2 = doc.createElement('y2')
y2.appendChild(doc.createTextNode(str((objects_axis[i][5]))))
bndbox.appendChild(y2)
x3 = doc.createElement('x3')
x3.appendChild(doc.createTextNode(str((objects_axis[i][6]))))
bndbox.appendChild(x3)
y3 = doc.createElement('y3')
y3.appendChild(doc.createTextNode(str((objects_axis[i][7]))))
bndbox.appendChild(y3)
f = open(save_path,'w')
f.write(doc.toprettyxml(indent = ''))
f.close()
class_list = ['ignored regions','pedestrian','people','bicycle','car','van','truck','tricycle','awning-tricycle','bus','motor','others']
def format_label(txt_list):
format_data = []
for i in txt_list[0:]:
format_data.append(
[int(xy) for xy in i.split(',')[:8]]
# {'x0': int(i.split(' ')[0]),
# 'x1': int(i.split(' ')[2]),
# 'x2': int(i.split(' ')[4]),
# 'x3': int(i.split(' ')[6]),
# 'y1': int(i.split(' ')[1]),
# 'y2': int(i.split(' ')[3]),
# 'y3': int(i.split(' ')[5]),
# 'y4': int(i.split(' ')[7]),
# 'class': class_list.index(i.split(' ')[8]) if i.split(' ')[8] in class_list else 0,
# 'difficulty': int(i.split(' ')[9])}
)
# if i.split(',')[8] not in class_list :
# print ('warning found a new label :', i.split(',')[8])
# exit()
return np.array(format_data)
def clip_image(file_idx, image, boxes_all, width, height, stride_w, stride_h):
if len(boxes_all) > 0:
shape = image.shape
for start_h in range(0, shape[0], stride_h):
for start_w in range(0, shape[1], stride_w):
boxes = copy.deepcopy(boxes_all)
box = np.zeros_like(boxes_all)
start_h_new = start_h
start_w_new = start_w
if start_h + height > shape[0]:
start_h_new = shape[0] - height
if start_w + width > shape[1]:
start_w_new = shape[1] - width
top_left_row = max(start_h_new, 0)
top_left_col = max(start_w_new, 0)
bottom_right_row = min(start_h + height, shape[0])
bottom_right_col = min(start_w + width, shape[1])
subImage = image[top_left_row:bottom_right_row, top_left_col: bottom_right_col]
box[:, 0] = boxes[:, 0]- top_left_col
box[:, 2] = boxes[:, 0] + boxes[:, 2]- top_left_col
box[:, 4] = boxes[:, 4]
box[:, 0] = [max(i, 0) for i in box[:, 0]] # 限制框的大小
# box[:, 6] = boxes[:, 6] - top_left_col
box[:, 1] = boxes[:, 1] - top_left_row
box[:, 3] = boxes[:, 1] + boxes[:, 3] - top_left_row
box[:, 5] = boxes[:, 5]
box[:, 1] = [max(i, 0) for i in box[:, 1]]
# box[:, 7] = boxes[:, 7] - top_left_row
# box[:, 8] = boxes[:, 8]
center_y = 0.5*(box[:, 1] + box[:, 3])
center_x = 0.5*(box[:, 0] + box[:, 2])
# print('center_y', center_y)
# print('center_x', center_x)
# print ('boxes', boxes)
# print ('boxes_all', boxes_all)
# print ('top_left_col', top_left_col, 'top_left_row', top_left_row)
cond1 = np.intersect1d(np.where(center_y[:]>=0)[0], np.where(center_x[:]>=0 )[0])
cond2 = np.intersect1d(np.where(center_y[:] <= (bottom_right_row - top_left_row))[0],
np.where(center_x[:] <= (bottom_right_col - top_left_col))[0])
idx = np.intersect1d(cond1, cond2)
# idx = np.where(center_y[:]>=0 and center_x[:]>=0 and center_y[:] <= (bottom_right_row - top_left_row) and center_x[:] <= (bottom_right_col - top_left_col))[0]
# save_path, im_width, im_height, objects_axis, label_name
if len(idx) > 0:
name="%s_%04d_%04d.jpg" % (file_idx, top_left_row, top_left_col)
print(name)
xml = os.path.join(save_dir, 'annotations_600_xml', "%s_%04d_%04d.xml" % (file_idx, top_left_row, top_left_col))
save_to_xml(xml, subImage.shape[1], subImage.shape[0], box[idx, :], class_list, str(name))
# save_to_txt(xml, box[idx, :])
# print ('save xml : ', xml)
if subImage.shape[0] > 5 and subImage.shape[1] >5:
img = os.path.join(save_dir, 'images_600', "%s_%04d_%04d.jpg" % (file_idx, top_left_row, top_left_col))
cv2.imwrite(img, subImage)
print ('class_list', len(class_list))
raw_data = 'D:/datasets/VisDrone/VisDrone2019-DET-val/'
raw_images_dir = os.path.join(raw_data, 'images')
raw_label_dir = os.path.join(raw_data, 'annotations')
save_dir = 'D:/datasets/VisDrone/VisDrone2019-DET-val/'
images = [i for i in os.listdir(raw_images_dir) if 'jpg' in i]
labels = [i for i in os.listdir(raw_label_dir) if 'txt' in i]
print ('find image', len(images))
print ('find label', len(labels))
min_length = 1e10
max_length = 1
img_h, img_w, stride_h, stride_w = 600, 600, 450, 450
for idx, img in enumerate(images):
# img = 'P1524.png'
img_data = misc.imread(os.path.join(raw_images_dir, img))
print (idx, 'read image', img)
# if len(img_data.shape) == 2:
# img_data = img_data[:, :, np.newaxis]
# print ('find gray image')
txt_data = open(os.path.join(raw_label_dir, img.replace('jpg', 'txt')), 'r').readlines()
# print (idx, len(format_label(txt_data)), img_data.shape)
# if max(img_data.shape[:2]) > max_length:
# max_length = max(img_data.shape[:2])
# if min(img_data.shape[:2]) < min_length:
# min_length = min(img_data.shape[:2])
# if idx % 50 ==0:
# print (idx, len(format_label(txt_data)), img_data.shape)
# print (idx, 'min_length', min_length, 'max_length', max_length)
box = format_label(txt_data)
# box = dele(box)
clip_image(img.strip('.jpg'), img_data, box, img_h, img_w, stride_h, stride_w)
# rm val/images/* && rm val/labeltxt/*
二、提取图片名字
extract_name.py
# P02 批量读取文件名（不带后缀）
import os
file_path = "D:/datasets/VisDrone/VisDrone2019-DET-val/annotations_600/"
path_list = os.listdir(file_path) # os.listdir(file)会历遍文件夹内的文件并返回一个列表
print(path_list)
path_name = [] # 把文件列表写入save.txt中
def saveList(pathName):
for file_name in pathName:
with open("name_600_val.txt", "a") as f:
f.write(file_name.split(".")[0] + "\n")
def dirList(path_list):
for i in range(0, len(path_list)):
path = os.path.join(file_path, path_list[i])
if os.path.isdir(path):
saveList(os.listdir(path))
dirList(path_list)
saveList(path_list)
三、将xml标注文件转成yolo需要的标注格式
xml2yolo.py
# 缺陷坐标xml转txt
import xml.etree.ElementTree as ET
import os
classes = ['ignored regions','pedestrian','people','bicycle','car','van','truck','tricycle','awning-tricycle','bus','motor','others'] # 输入缺陷名称，必须与xml标注名称一致
train_file = 'images_val_600_test.txt'
train_file_txt = ''
wd = os.getcwd()
def convert(size, box):
dw = 1. / size[0]
dh = 1. / size[1]
box = list(box)
box[1] = min(box[1], size[0]) # 限制目标的范围在图片尺寸内
box[3] = min(box[3], size[1])
x = ((box[0] + box[1]) / 2.0) * dw
y = ((box[2] + box[3]) / 2.0) * dh
w = (box[1] - box[0]) * dw
h = (box[3] - box[2]) * dh
return (x, y, w, h)
def convert_annotation(image_id):
in_file = open('D:/datasets/VisDrone/VisDrone2019-DET-val/annotations_600_xml/%s.xml' % (image_id)) # 读取xml文件路径
out_file = open('D:/datasets/VisDrone/labels_val_600/%s.txt' % (image_id), 'w') # 需要保存的txt格式文件路径
tree = ET.parse(in_file)
root = tree.getroot()
size = root.find('size')
w = int(size.find('width').text)
h = int(size.find('height').text)
for obj in root.iter('object'):
cls = obj.find('name').text
if cls not in classes: # 检索xml中的缺陷名称
continue
cls_id = classes.index(cls)
# import pdb
# pdb.set_trace()
if cls_id == 0 or cls_id == 11:
continue
xmlbox = obj.find('bndbox')
b = (float(xmlbox.find('xmin').text), float(xmlbox.find('xmax').text), float(xmlbox.find('ymin').text),
float(xmlbox.find('ymax').text))
bb = convert((w, h), b)
out_file.write(str(cls_id - 1) + " " + " ".join([str(a) for a in bb]) + '\n')
image_ids_train = open('D:/datasets/VisDrone/name_600_val.txt').read().strip().split() # 读取xml文件名索引
for image_id in image_ids_train:
convert_annotation(image_id)
anns = os.listdir('./VisDrone2019-DET-val/annotations_600_xml/')
for ann in anns:
ans = ''
outpath = wd + '/labels_val_600/' + ann
if ann[-3:] != 'xml':
continue
train_file_txt = train_file_txt + wd + '/VisDrone2019-DET-val' + '/images_600/' + ann[:-3] + 'jpg\n'
with open(train_file, 'w') as outfile:
outfile.write(train_file_txt)
四、Visdrone直接转成yolo格式
trans_yolo.py
import os
from pathlib import Path
from PIL import Image
import csv
def convert(size, box):
dw = 1. / size[0]
dh = 1. / size[1]
x = (box[0] + box[2] / 2) * dw
y = (box[1] + box[3] / 2) * dh
w = box[2] * dw
h = box[3] * dh
return (x, y, w, h)
wd = os.getcwd()
if not os.path.exists('labels_val'):
os.makedirs('labels_val')
train_file = 'images_val.txt'
train_file_txt = ''
anns = os.listdir('./VisDrone2019-DET-val/annotations')
for ann in anns:
ans = ''
outpath = wd + '/labels_val/' + ann
if ann[-3:] != 'txt':
continue
with Image.open(wd + './VisDrone2019-DET-val/images/' + ann[:-3] + 'jpg') as Img:
img_size = Img.size
with open(wd + './VisDrone2019-DET-val/annotations/' + ann, newline='') as csvfile:
spamreader = csv.reader(csvfile)
# import pdb
# pdb.set_trace()
for row in spamreader:
if row[4] == '0':
continue
bb = convert(img_size, tuple(map(int, row[:4])))
ans = ans + str(int(row[5])-1) + ' ' + ' '.join(str(a) for a in bb) + '\n'
with open(outpath, 'w') as outfile:
outfile.write(ans)
train_file_txt = train_file_txt + wd + '/images/' + ann[:-3] + 'jpg\n'
with open(train_file, 'w') as outfile:
outfile.write(train_file_txt)
五、使用xml格式画框
draw_visdrone.py
import os
import os.path
import xml.etree.cElementTree as ET
import cv2
def draw(image_path, xml_path, root_saved_path):
""" 图片根据标注画框 """
src_img_path = image_path
src_ann_path = xml_path
for file in os.listdir(src_ann_path):
# print(file)
file_name, suffix = os.path.splitext(file)
# import pdb
# pdb.set_trace()
if suffix == '.xml':
# print(file)
xml_path = os.path.join(src_ann_path, file)
image_path = os.path.join(src_img_path, file_name+'.jpg')
img = cv2.imread(image_path)
tree = ET.parse(xml_path)
root = tree.getroot()
# import pdb
# pdb.set_trace()
for obj in root.iter('object'):
name = obj.find('name').text
xml_box = obj.find('bndbox')
x1 = int(xml_box.find('xmin').text)
x2 = int(xml_box.find('xmax').text)
y1 = int(xml_box.find('ymin').text)
y2 = int(xml_box.find('ymax').text)
cv2.rectangle(img, (x1, y1), (x2, y2), (255, 0, 0), thickness=2)
# 字为绿色
# cv2.putText(img, name, (x1, y1), cv2.FONT_HERSHEY_COMPLEX, 0.7, (0, 255, 0), thickness=2)
cv2.imwrite(os.path.join(root_saved_path, file_name+'.jpg'), img)
if __name__ == '__main__':
image_path = "D:/datasets/VisDrone/VisDrone2019-DET-train/images_600"
xml_path = "D:/datasets/VisDrone/VisDrone2019-DET-train/annotations_600"
root_saved_path = "D:/datasets/VisDrone/VisDrone2019-DET-train/result"
draw(image_path, xml_path, root_saved_path)
六、将txt格式转换成xml格式
txt2xml.py
# coding: utf-8
# author: HXY
# 2020-4-17
""" 该脚本用于visdrone数据处理； 将annatations文件夹中的txt标签文件转换为XML文件； txt标签内容为： <bbox_left>,<bbox_top>,<bbox_width>,<bbox_height>,<score>,<object_category>,<truncation>,<occlusion> 类别： ignored regions(0), pedestrian(1), people(2), bicycle(3), car(4), van(5), truck(6), tricycle(7), awning-tricycle(8), bus(9), motor(10), others(11) """
import os
import cv2
import time
from xml.dom import minidom
name_dict = {
'0': 'ignored regions', '1': 'pedestrian', '2': 'people',
'3': 'bicycle', '4': 'car', '5': 'van', '6': 'truck',
'7': 'tricycle', '8': 'awning-tricycle', '9': 'bus',
'10': 'motor', '11': 'others'}
def transfer_to_xml(pic, txt, file_name):
xml_save_path = 'D:/datasets/VisDrone/VisDrone2019-DET-val/annotations_xml' # 生成的xml文件存储的文件夹
if not os.path.exists(xml_save_path):
os.mkdir(xml_save_path)
img = cv2.imread(pic)
img_w = img.shape[1]
img_h = img.shape[0]
img_d = img.shape[2]
doc = minidom.Document()
annotation = doc.createElement("annotation")
doc.appendChild(annotation)
folder = doc.createElement('folder')
folder.appendChild(doc.createTextNode('visdrone'))
annotation.appendChild(folder)
filename = doc.createElement('filename')
filename.appendChild(doc.createTextNode(file_name))
annotation.appendChild(filename)
source = doc.createElement('source')
database = doc.createElement('database')
database.appendChild(doc.createTextNode("Unknown"))
source.appendChild(database)
annotation.appendChild(source)
size = doc.createElement('size')
width = doc.createElement('width')
width.appendChild(doc.createTextNode(str(img_w)))
size.appendChild(width)
height = doc.createElement('height')
height.appendChild(doc.createTextNode(str(img_h)))
size.appendChild(height)
depth = doc.createElement('depth')
depth.appendChild(doc.createTextNode(str(img_d)))
size.appendChild(depth)
annotation.appendChild(size)
segmented = doc.createElement('segmented')
segmented.appendChild(doc.createTextNode("0"))
annotation.appendChild(segmented)
with open(txt, 'r') as f:
lines = [f.readlines()]
for line in lines:
for boxes in line:
box = boxes.strip('/n')
box = box.split(',')
x_min = box[0]
y_min = box[1]
x_max = int(box[0]) + int(box[2])
y_max = int(box[1]) + int(box[3])
object_name = name_dict[box[5]]
# if object_name is 'ignored regions' or 'others':
# continue
object = doc.createElement('object')
nm = doc.createElement('name')
nm.appendChild(doc.createTextNode(object_name))
object.appendChild(nm)
pose = doc.createElement('pose')
pose.appendChild(doc.createTextNode("Unspecified"))
object.appendChild(pose)
truncated = doc.createElement('truncated')
truncated.appendChild(doc.createTextNode("1"))
object.appendChild(truncated)
difficult = doc.createElement('difficult')
difficult.appendChild(doc.createTextNode("0"))
object.appendChild(difficult)
bndbox = doc.createElement('bndbox')
xmin = doc.createElement('xmin')
xmin.appendChild(doc.createTextNode(x_min))
bndbox.appendChild(xmin)
ymin = doc.createElement('ymin')
ymin.appendChild(doc.createTextNode(y_min))
bndbox.appendChild(ymin)
xmax = doc.createElement('xmax')
xmax.appendChild(doc.createTextNode(str(x_max)))
bndbox.appendChild(xmax)
ymax = doc.createElement('ymax')
ymax.appendChild(doc.createTextNode(str(y_max)))
bndbox.appendChild(ymax)
object.appendChild(bndbox)
annotation.appendChild(object)
with open(os.path.join(xml_save_path, file_name + '.xml'), 'w') as x:
x.write(doc.toprettyxml())
x.close()
f.close()
if __name__ == '__main__':
t = time.time()
print('Transfer .txt to .xml...ing....')
txt_folder = 'D:/datasets/VisDrone/VisDrone2019-DET-val/annotations' # visdrone txt标签文件夹
txt_file = os.listdir(txt_folder)
img_folder = 'D:/datasets/VisDrone/VisDrone2019-DET-val/images' # visdrone 照片所在文件夹
for txt in txt_file:
txt_full_path = os.path.join(txt_folder, txt)
img_full_path = os.path.join(img_folder, txt.split('.')[0] + '.jpg')
try:
transfer_to_xml(img_full_path, txt_full_path, txt.split('.')[0])
except Exception as e:
print(e)
print("Transfer .txt to .XML sucessed. costed: {:.3f}s...".format(time.time() - t))
七、将原始图像变成二值图像（目标存在是白色，背景是黑色）
img2binary.py
# -*- coding:utf-8 -*-
# 把标注目标映射为二值图像
import matplotlib.pyplot as plt
import cv2, os
from xml.dom import minidom
import xml.etree.ElementTree as ET
import numpy as np
def tobinary(img_path):
img_list = os.listdir(img_path)
for img_name_id in img_list:
# import pdb
# pdb.set_trace()
img_id, _ = os.path.splitext(img_name_id)
img_file = os.path.join(img_path, img_id + ".jpg")
image = cv2.imread(img_file)
xml_file = os.path.join(xml_path, img_id + ".xml")
weight, height = image.shape[0], image.shape[1]
bg = np.zeros((weight, height), dtype = np.uint8)
tree = ET.parse(xml_file)
root = tree.getroot()
for obj in root.iter("object"):
name = obj.find("name").text
xmlbox = obj.find("bndbox")
xmin = int(xmlbox.find('xmin').text)
ymin = int(xmlbox.find('ymin').text)
xmax = int(xmlbox.find('xmax').text)
ymax = int(xmlbox.find('ymax').text)
color = (255, 255, 255)
cv2.rectangle(bg, (xmin, ymin), (xmax, ymax), color, -1)
img_name = img_id + ".jpg"
out_file = os.path.join(out_path, img_name)
cv2.imwrite(out_file, bg)
if __name__ == "__main__":
img_path = "D:/datasets/VisDrone/VisDrone2019-DET-train/images/"
xml_path = "D:/datasets/VisDrone/VisDrone2019-DET-train/annotations_xml/"
out_path = "D:/datasets/VisDrone/VisDrone2019-DET-train/output/"
tobinary(img_path)
# xml = open("D:/datasets/VisDrone/VisDrone2019-DET-train/annotations_xml/0000001_02999_d_0000005.xml")
# import pdb
# pdb.set_trace()
八、绘制PR曲线
draw.py
from pr import *
clas= ['pedestrian','people','bicycle','car','van','truck',
'tricycle','awning-tricycle','bus','motor'] # 类别
visdrone_file=["yolov5.txt", "centernet.txt", "ucgnet.txt"] # 放在Prediction里面"Faster_RCNN.txt"
visdrone_algorithm=["yolov5", "centernet", "ucgnet"] # 对应上面的文件的算法名称
use_07_metric=[False, False, False]
ground_truth_file=["visdrone_gt.txt", "visdrone_gt.txt", "visdrone_gt.txt"] # 放在Ground Truth里面
# DOTA_file=["MKD-Net-128.txt"] # 放在Prediction里面"Faster_RCNN.txt"
# DOTA_algorithm=["MKD-Net-128"] # 对应上面的文件的算法名称
# use_07_metric=[False]
# ground_truth_file=["dota_gt.txt"] # 放在Ground Truth里面
# 绘制出了各个类别的Precision-Recall曲线
draw_pr(visdrone_file,visdrone_algorithm,ground_truth_file,clas,use_07_metric)
pr.py
# -*- coding: UTF-8 -*-
#以faster和yolo为例
import numpy as np
import math
import matplotlib.pyplot as plt
def get_pr_data_map(prediction_file,ground_truth_file,cls_name,use_07_metric,ovthresh=0.4):
with open(ground_truth_file, 'r') as f:####读取各个算法的gt.txt文件
lines_gt = f.readlines() #gt.txt的每一行，line_gt为一维数组
with open(prediction_file, 'r') as f:####读取各个算法的pre.txt文件
lines_pre = f.readlines() #读取pre.txt的每一行，lines_gt为一维数组
#根据txt文件中每一行的空格进行划分，splitlines_gt为二维数组，行为每一行的数据，列为每一行的数据划分
splitlines_gt = [x.strip().split(' ') for x in lines_gt]
#imagenames_gt为一维数组，当第二列的类别与遍历的类别数相同时，将第一列加入到imagenames_gt中
imagenames_gt = [ x[0] for x in splitlines_gt ]
#print(imagenames_gt)
class_recs = {
}
# #BB_gt为二维数组，当第二列的类别与遍历的类别数相同时，将第二列之后的加入到BB_gt中
BB_gt=np.array([[math.ceil(float(z)) for z in x[2:]] for x in splitlines_gt])
for i in range(len(imagenames_gt)):
#如果类别中没有该遍历的类别，则更新
if imagenames_gt[i] not in class_recs:
class_recs.update({
imagenames_gt[i]:{
"bbox":[],"det":[],"difficult":[]}})
#否则将BB_gt[i]插入到类别框信息中
class_recs[imagenames_gt[i]]["bbox"].append(BB_gt[i])
class_recs[imagenames_gt[i]]["det"].append(False)
class_recs[imagenames_gt[i]]["difficult"].append(False)
#npos为imagenames_gt的长度
npos=len(imagenames_gt)
splitlines_pre = [x.strip().split(' ') for x in lines_pre]
#image_ids为一维数组，当第二列的类别与遍历的类别数相同时，将第一列加入到image_ids中
image_ids = [x[0] for x in splitlines_pre]
#提取该类别预测框的置信度，当第二列的类别与遍历的类别数相同时，将第三列加入到confidence中
confidence = np.array([float(x[2]) for x in splitlines_pre])
#提取该类别预测框的bounging-boxes（二维数组），当第二列的类别与遍历的类别数相同时,将第三列之后的BBOX坐标插入
BB_pre = np.array([[float(z) for z in x[3:]] for x in splitlines_pre])
#按置信度大小将其索引从小到大排序（生成有顺序的一维数组）
sorted_ind = np.argsort(-confidence)
#按置信度大小将置信度从小到大排序（生成有顺序的一维数组）
sorted_scores = np.sort(-confidence)
#根据索引排序相应的bbox的坐标值（生成按置信度大小排列的二维数组）
BB_pre = BB_pre[sorted_ind, :]
#按置信度大小重新排列image_ids
image_ids = [image_ids[x] for x in sorted_ind]
nd = len(image_ids)
tp = np.zeros(nd)
fp = np.zeros(nd)
pr_data_map=[]
# print(nd)###############################################################################################################
for d in range(nd):
#按置信度顺序提取类别框
#print(image_ids[d].get(image_ids[d]))
# if class_recs[image_ids[d]] not in class_recs:
# print(class_recs)
if class_recs.get(image_ids[d]):
#print(image_ids[d])
R = class_recs[image_ids[d]]
print(R)
bb = np.array(BB_pre[d, :]).astype(float)#按置信度顺序提取bounding box
ovmax = -np.inf#ovmax为负无穷大的数
BBGT = np.array(R['bbox']).astype(float)#按置信度顺序提取groudtruth bbox
#计算iou
if BBGT.size > 0:
ixmin = np.maximum(BBGT[:, 0], bb[0])
iymin = np.maximum(BBGT[:, 1], bb[1])
ixmax = np.minimum(BBGT[:, 2], bb[2])
iymax = np.minimum(BBGT[:, 3], bb[3])
iw = np.maximum(ixmax - ixmin + 1., 0.)
ih = np.maximum(iymax - iymin + 1., 0.)
inters = iw * ih
uni = ((bb[2] - bb[0] + 1.) * (bb[3] - bb[1] + 1.) +
(BBGT[:, 2] - BBGT[:, 0] + 1.) *
(BBGT[:, 3] - BBGT[:, 1] + 1.) - inters)
overlaps = inters / uni#重叠率
ovmax = np.max(overlaps)#按重叠率的大小从大到小排序重叠率
jmax = np.argmax(overlaps)#根据重叠率大小重新排序的索引
if ovmax > ovthresh:#ovthresh=0.5阈值为0.5，判断tp和fp
if not R['difficult'][jmax]:
if not R['det'][jmax]:
tp[d] = 1.
print(tp)
R['det'][jmax] = 1
else:
fp[d] = 1.
else:
fp[d] = 1.
else:
#print("*************************************************************"+str(d))
fp[d]=1.
fp = np.cumsum(fp)
tp = np.cumsum(tp)
#rec=tp/正样本数
rec = tp / float(npos)
print(len(rec))
# print(type(rec))
#perc=tp/(tp+fp)
prec = tp / np.maximum(tp + fp, np.finfo(np.float64).eps)
print(len(prec))
# print(type(prec))
ap = voc_ap(rec, prec, use_07_metric)
print(ap)
###################################################
# mAP_rec+=rec
# print("mAP_rec:"+str(mAP_rec)+"\n")
# mAP_prec+=prec
# print("mAP_prec:"+str(mAP_prec)+"\n")
# mAP+=ap
# print("mAP:"+str(mAP)+"\n")
###################################################
# if not pr_data.has_key(cls_name[cls_num]):
# rec=np.array(rec)
# prec=np.array(prec)
# ap=np.array(ap)
pr_data_map=np.array([rec.tolist(),prec.tolist()])#,ap.tolist()]
pr_data_map=pr_data_map.tolist()
#print(pr_data_map)
return pr_data_map
def get_pr_data(prediction_file,ground_truth_file,cls_name,use_07_metric,ovthresh=0.5):
with open(ground_truth_file, 'r') as f:####读取各个算法的gt.txt文件
lines_gt = f.readlines() #gt.txt的每一行，line_gt为一维数组
with open(prediction_file, 'r') as f:####读取各个算法的pre.txt文件
lines_pre = f.readlines() #读取pre.txt的每一行，lines_gt为一维数组
pr_data={
}
for cls_num in range(len(cls_name)): #遍历每个类别数
#ground_truth
#根据txt文件中每一行的空格进行划分，splitlines_gt为二维数组，行为每一行的数据，列为每一行的数据划分
splitlines_gt = [x.strip().split(' ') for x in lines_gt]
# print(splitlines_gt)
#imagenames_gt为一维数组，当第二列的类别与遍历的类别数相同时，将第一列加入到imagenames_gt中
# imagenames_gt = []
# for x in splitlines_gt:
# if int(x[1]) - 1 == cls_num:
# imagenames_gt = x[0]
imagenames_gt = [ x[0] for x in splitlines_gt if int(x[1]) - 1 ==cls_num]
# print(imagenames_gt)
#BB_gt为二维数组，当第二列的类别与遍历的类别数相同时，将第二列之后的加入到BB_gt中
BB_gt=np.array([[math.ceil(float(z)) for z in x[2:]] for x in splitlines_gt if int(x[1]) - 1 ==cls_num])
#创建class_recs数组
class_recs = {
}
for i in range(len(imagenames_gt)):
# import pdb
# pdb.set_trace()
#如果类别中没有该遍历的类别，则更新
if imagenames_gt[i] not in class_recs:
class_recs.update({
imagenames_gt[i]:{
"bbox":[],"det":[],"difficult":[]}})
#否则将BB_gt[i]插入到类别框信息中
class_recs[imagenames_gt[i]]["bbox"].append(BB_gt[i])
class_recs[imagenames_gt[i]]["det"].append(False)
class_recs[imagenames_gt[i]]["difficult"].append(False)
# print(class_recs[imagenames_gt])
#npos为imagenames_gt的长度
npos=len(imagenames_gt)
# print("##################")
# print(class_recs)
#prediction
#对预测框的每一行根据空格划分，splitlines_pre为二维数组
splitlines_pre = [x.strip().split(' ') for x in lines_pre]
# print(splitlines_pre)
#image_ids为一维数组，当第二列的类别与遍历的类别数相同时，将第一列加入到image_ids中
image_ids = [ x[0] for x in splitlines_pre if int(x[1]) - 1 ==cls_num]
# print(cls_num)
#提取该类别预测框的置信度，当第二列的类别与遍历的类别数相同时，将第三列加入到confidence中
confidence = np.array([float(x[2]) for x in splitlines_pre if int(x[1]) - 1 ==cls_num])
# print("********", confidence)
#提取该类别预测框的bounging-boxes（二维数组），当第二列的类别与遍历的类别数相同时,将第三列之后的BBOX坐标插入
BB_pre = np.array([[float(z) for z in x[3:]] for x in splitlines_pre if int(x[1]) - 1 ==cls_num])
#按置信度大小将其索引从小到大排序（生成有顺序的一维数组）
sorted_ind = np.argsort(-confidence)
# print(sorted_ind,"*************")
#按置信度大小将置信度从小到大排序（生成有顺序的一维数组）
sorted_scores = np.sort(-confidence)
#根据索引排序相应的bbox的坐标值（生成按置信度大小排列的二维数组）
BB_pre = BB_pre[sorted_ind, :]
#按置信度大小重新排列image_ids
image_ids = [image_ids[x] for x in sorted_ind]
#该类别的预测框的数量
nd = len(image_ids)
tp = np.zeros(nd)#将长度为nd数组置0
fp = np.zeros(nd)
# print(nd)###############################################################################################################
for d in range(nd):
#按置信度顺序提取类别框
#print(image_ids[d].get(image_ids[d]))
# if class_recs[image_ids[d]] not in class_recs:
if class_recs.get(image_ids[d]):
# print(image_ids[d])
R = class_recs[image_ids[d]]
#print(R)
bb = np.array(BB_pre[d, :]).astype(float)#按置信度顺序提取bounding box
ovmax = -np.inf#ovmax为负无穷大的数
BBGT = np.array(R['bbox']).astype(float)#按置信度顺序提取groudtruth bbox
#计算iou
if BBGT.size > 0:
ixmin = np.maximum(BBGT[:, 0], bb[0])
iymin = np.maximum(BBGT[:, 1], bb[1])
ixmax = np.minimum(BBGT[:, 2], bb[2])
iymax = np.minimum(BBGT[:, 3], bb[3])
# ixmin = max(BBGT[:, 0], bb[0])
# iymin = max(BBGT[:, 1], bb[1])
# ixmax = min(BBGT[:, 2], bb[2])
# iymax = min(BBGT[:, 3], bb[3])
# import pdb
# pdb.set_trace()
# if ixmin >= ixmax or iymax <= iymin:
# return 0
# else:
# S1 = (BBGT[2]-BBGT[0])*(BBGT[3]-BBGT[1])
# S2 = (bb[2]-bb[0])*(bb[3]-bb[1])
iw = np.maximum(ixmax - ixmin + 1., 0.)
ih = np.maximum(iymax - iymin + 1., 0.)
inters = iw * ih
uni = ((bb[2] - bb[0] + 1.) * (bb[3] - bb[1] + 1.) +
(BBGT[:, 2] - BBGT[:, 0] + 1.) *
(BBGT[:, 3] - BBGT[:, 1] + 1.) - inters)
overlaps = inters / uni#重叠率
ovmax = np.max(overlaps)#按重叠率的大小从大到小排序重叠率
jmax = np.argmax(overlaps)#根据重叠率大小重新排序的索引
if ovmax > ovthresh:#ovthresh=0.5阈值为0.5，判断tp和fp
if not R['difficult'][jmax]:
if not R['det'][jmax]:
tp[d] = 1.
R['det'][jmax] = 1
else:
fp[d] = 1.
else:
fp[d] = 1.
else:
#print("*************************************************************"+str(d))
fp[d]=1.
# import pdb
# pdb.set_trace()
print(cls_name[cls_num], np.sum(tp), np.sum(fp), npos)
fp = np.cumsum(fp)
tp = np.cumsum(tp)
#rec=tp/正样本数
rec = tp / float(npos)
#print(rec)
# print(type(rec))
#perc=tp/(tp+fp)
prec = tp / np.maximum(tp + fp, np.finfo(np.float64).eps)
#?print(prec)
# print(type(prec))
ap = voc_ap(rec, prec, use_07_metric)
#print(ap)
###################################################
# mAP_rec+=rec
# print("mAP_rec:"+str(mAP_rec)+"\n")
# mAP_prec+=prec
# print("mAP_prec:"+str(mAP_prec)+"\n")
# mAP+=ap
# print("mAP:"+str(mAP)+"\n")
###################################################
if cls_name[cls_num] not in pr_data:
pr_data.update({
cls_name[cls_num]:[rec,prec,ap]})
return pr_data #,mAP/7,np.array(mAP_rec/7),np.array(mAP_prec/7)
def voc_ap(rec, prec, use_07_metric):#由于use_07_metric=true时计算结果于实际更接近
#计算ap，use_07_metric=true，# 2010年以前按recall等间隔取11个不同点处的精度值做平均(0., 0.1, 0.2, …, 0.9, 1.0)
if use_07_metric:
ap = 0.
for t in np.arange(0., 1.1, 0.1):#（[0.0,0.1,0.2,0.3,...,1.0])
#print(11111111111111111111111111)
if np.sum(rec >= t) == 0:
p = 0
else:
p = np.max(prec[rec >= t])
ap = ap + p / 11.
#use_07_metric=false # 2010年以后取所有不同的recall对应的点处的精度值做平均
else:
mrec = np.concatenate(([0.], rec, [1.]))
mpre = np.concatenate(([0.], prec, [0.]))
for i in range(mpre.size - 1, 0, -1):
mpre[i - 1] = np.maximum(mpre[i - 1], mpre[i])
i = np.where(mrec[1:] != mrec[:-1])[0]
ap = np.sum((mrec[i + 1] - mrec[i]) * mpre[i + 1])
return ap
def draw_pr(prediction_file,prediction_algorithm,ground_truth_file,cls,use_07_metric):
# for i in range(len(prediction_file)):
# pr_data=get_pr_data("Prediction/"+prediction_file[i],"Ground_Truth/"+ground_truth_file,cls)
# for cls_name in pr_data:
# plt.plot(pr_data[cls_name][0],pr_data[cls_name][1],label=prediction_algorithm[i]+' mAP='+str(round(pr_data[cls_name][2],3)))
# title='PR Curve of '+cls_name
# plt.title(title)
# plt.xlabel('Recall')
# plt.ylabel('Precision')
# plt.ylim([0.0, 1.0])
# plt.xlim([0.0, 1.0])
# plt.grid(ls='-.')
# plt.legend()
# plt.savefig("Images/"+title+'.png', dpi=300)
# plt.show()
# plt.close()
# for i in range(len(prediction_file)):
# pr_data=get_pr_data("Prediction/"+prediction_file[i],"Ground_Truth/"+ground_truth_file,cls)
#画多个算法多个类别的pr曲线
MAP=[0,0,0,0,0,0]#用于统计map
for cls_name in cls:
for i in range(len(prediction_file)):
pr_data=get_pr_data("Prediction/"+prediction_file[i],"Ground_Truth/"+ground_truth_file[i],cls,use_07_metric[i])
# import pdb
# pdb.set_trace()
MAP[i]=MAP[i]+pr_data[cls_name][2]
# print(pr_data)
if i==0:
plt.plot(pr_data[cls_name][0],pr_data[cls_name][1],label=prediction_algorithm[i]+' AP='+str(round(pr_data[cls_name][2],3)),color="#054E9F")
elif i==1:
plt.plot(pr_data[cls_name][0],pr_data[cls_name][1],label=prediction_algorithm[i]+' AP='+str(round(pr_data[cls_name][2],3)),color="#FFA500")
elif i==2:
plt.plot(pr_data[cls_name][0],pr_data[cls_name][1],label=prediction_algorithm[i]+' AP='+str(round(pr_data[cls_name][2],3)),color="#B0C4DE")
elif i==3:
plt.plot(pr_data[cls_name][0],pr_data[cls_name][1],label=prediction_algorithm[i]+' AP='+str(round(pr_data[cls_name][2],3)),color="#008000")
elif i==4:
plt.plot(pr_data[cls_name][0],pr_data[cls_name][1],label=prediction_algorithm[i]+' AP='+str(round(pr_data[cls_name][2],3)),color="#BA55D3")
elif i==5:
plt.plot(pr_data[cls_name][0],pr_data[cls_name][1],label=prediction_algorithm[i]+' AP='+str(round(pr_data[cls_name][2],3)),color="#FF0000")
# plt.plot(pr_data[cls_name][0],pr_data[cls_name][1],label=prediction_algorithm[i]+' AP='+str(round(pr_data[cls_name][2],3)),color="#054E9F")
title=cls_name+' PR Curve'
plt.title(title,fontsize=15)
plt.xlabel('Recall',fontsize=15)
plt.ylabel('Precision',fontsize=15)
plt.ylim([0.0, 1.0])
plt.xlim([0.0, 1.0])
plt.grid(ls='-.')
plt.legend()
plt.savefig("Image_PR/"+title+'.png', dpi=600)
print("save....ok!!!")
plt.show()
plt.close()
#画各个算法map的pr曲线
# for i in range(len(prediction_file)):
# pr_data_map=get_pr_data_map("Prediction/"+prediction_file[i],"Ground_Truth/"+ground_truth_file[i],cls,use_07_metric[i])
# if i==0:
# plt.plot(pr_data_map[0],pr_data_map[1],label=prediction_algorithm[i]+' mAP='+str(round(MAP[i]/len(cls),3)),color="#054E9F")
# elif i==1:
# plt.plot(pr_data_map[0],pr_data_map[1],label=prediction_algorithm[i]+' mAP='+str(round(MAP[i]/len(cls),3)),color="#FFA500")
# elif i==2:
# plt.plot(pr_data_map[0],pr_data_map[1],label=prediction_algorithm[i]+' mAP='+str(round(MAP[i]/len(cls),3)),color="#B0C4DE")
# elif i==3:
# plt.plot(pr_data_map[0],pr_data_map[1],label=prediction_algorithm[i]+' mAP='+str(round(MAP[i]/len(cls),3)),color="#008000")
# elif i==4:
# plt.plot(pr_data_map[0],pr_data_map[1],label=prediction_algorithm[i]+' mAP='+str(round(MAP[i]/len(cls),3)),color="#BA55D3")
# elif i==5:
# plt.plot(pr_data_map[0],pr_data_map[1],label=prediction_algorithm[i]+' mAP='+str(round(MAP[i]/len(cls),3)),color="#FF0000")
# #plt.plot(pr_data_map[0],pr_data_map[1],label=prediction_algorithm[i]+' mAP='+str(round(MAP[i]/len(cls),3)),color="#054E9F")
# title='Precision-Recall Curve'
# plt.title(title,fontsize=15)
# plt.xlabel('Recall',fontsize=15)
# plt.ylabel('Precision',fontsize=15)
# plt.ylim([0.0, 1.0])
# plt.xlim([0.0, 1.0])
# plt.grid(ls='-.')
# plt.legend()
# plt.savefig("Image_PR/"+title+'.png', dpi=600)
# print("save....ok!!!")
# plt.show()
# plt.close()
今天的文章Visdrone数据集 | 数据前&后处理操作系列分享到此就结束了，感谢您的阅读。
版权声明：本文内容由互联网用户自发贡献，该文观点仅代表作者本人。本站仅提供信息存储空间服务，不拥有所有权，不承担相关法律责任。如发现本站有涉嫌侵权/违法违规的内容， 请发送邮件至 举报，一经查实，本站将立刻删除。
如需转载请保留出处：https://bianchenghao.cn/61918.html
