def analysis_xmls(dir_xml, ext, file_names):
dict_obj = {}
for file_name in file_names:
file_path = dir_xml + '/' + file_name + '.xml'
print(file_path)
pvr_obj = PascalVocReader(file_path)
shapes = pvr_obj.getShapes()
im_info = pvr_obj.get_imageInfo()
im_w = im_info[0][0]
im_h = im_info[0][1]
# im_c = im_info[0][2]
for shape in shapes:
key = shape[0]
left = shape[1][0]
top = shape[1][1]
right = shape[1][2]
bottom = shape[1][3]
if 1 >= left or 1 >= top:
print('error 0 !')
continue
if right >= (im_w - 2) or bottom >= (im_h - 2):
print('error 1 !')
continue
if left >= right or top >= bottom:
print('error 2 !')
continue
if key not in dict_obj:
dict_obj[key] = 1
else:
dict_obj[key] += 1
return dict_obj
def calculate_images(self, lis=None):
if lis == None:
lis = self.lis
img_count = {}
for file in lis:
xml_name = self.anno_path + '/' + file + '.xml'
if os.path.exists(xml_name):
pr = PR(xml_name)
shapes = pr.getShapes()
#
label_list = [s[0] for s in shapes]
else:
print('file not exist')
# else:
# lis.remove(file)
# continue
for k in self.label_count.keys():
if k in label_list:
try:
img_count[k] += 1
except:
img_count[k] = 1
return img_count, lis
def del_label(self, label_del):
flist = os.listdir(self.anno_path)
for filename in flist:
filepath = self.anno_path + '/' + filename
pr = PR(filepath)
shapes = pr.getShapes()
sl = pr.getSources()
localpath = os.path.join(self.im_path, sl['filename'])
pw = PW(sl['folder'],
sl['filename'],
sl['size'],
databaseSrc='Unknown',
localImgPath=localpath)
pw.verified = True
for shape in shapes:
if shape[0] != label_del:
label = shape[0]
difficult = 0
bndbox = shape[1]
pw.addBndBox(bndbox[0][0], bndbox[0][1],
bndbox[2][0], bndbox[2][1], label, difficult)
pw.save(filepath)
def label_filter(self, labels, new_dir):
flist = []
for lb in labels:
stemp = self.find_label(lb)
for fname in stemp:
if (fname in flist) == False:
flist.append(fname)
for filename in flist:
filepath = self.anno_path + '/' + filename + '.xml'
pr = PR(filepath)
shapes = pr.getShapes()
sl = pr.getSources()
localpath = os.path.join(self.im_path, sl['filename'])
pw = PW(sl['folder'],
sl['filename'],
sl['size'],
databaseSrc='Unknown',
localImgPath=localpath)
pw.verified = True
for shape in shapes:
if shape[0] in labels:
label = shape[0]
difficult = 0
bndbox = shape[1]
pw.addBndBox(bndbox[0][0], bndbox[0][1],
bndbox[2][0], bndbox[2][1], label, difficult)
pw.save(new_dir + '/' + filename + '.xml')
def model_test(net, test_list, input_dir, iter_num=1):
total_score = 0.0
score = 0.0
class_score = {}
for im_name in test_list:
if DEBUG:
print "############## scoring {0} ##############".format(im_name)
im_file = os.path.join(input_dir, 'img', im_name)
label_file = os.path.join(input_dir, 'annotation', im_name)
index = label_file.rindex('.')
label_file = label_file[:index] + ".xml"
if not os.path.isfile(label_file):
sys.stderr.write("can't find label file for {0}".format(im_file))
continue
label_reader = PascalVocReader(label_file)
size = label_reader.getSize()
shapes = label_reader.getShapes()
true_label = convert_shapes(shapes)
pre_label = predict_label(net, im_file)
t_s, s = compare_label(true_label, pre_label, class_score)
print "scoring {0}".format(im_name), t_s, s, s / t_s
total_score += t_s
score += s
print "iter {0} total_score".format(
iter_num), score, total_score, score / total_score
for i in class_score.iterkeys():
print "iter {0} class_score:".format(
iter_num), i, class_score[i], class_score[i][0] / class_score[i][1]
return total_score, score
def test_upper(self):
dir_name = os.path.abspath(os.path.dirname(__file__))
libs_path = os.path.join(dir_name, '..', 'libs')
sys.path.insert(0, libs_path)
from pascal_voc_io import PascalVocWriter
from pascal_voc_io import PascalVocReader
# Test Write/Read
writer = PascalVocWriter('tests',
'test', (512, 512, 1),
localImgPath='tests/test.bmp')
writer.addBndBox(60, 40, 430, 504, 'person', dict(difficult=1))
writer.addBndBox(113, 40, 450, 403, 'face', dict(difficult=1))
writer.save('tests/test.xml')
reader = PascalVocReader('tests/test.xml')
shapes = reader.getShapes()
personBndBox = shapes[0]
face = shapes[1]
self.assertEqual(personBndBox[0], 'person')
self.assertEqual(personBndBox[1], [(60, 40), (430, 40), (430, 504),
(60, 504)])
self.assertEqual(face[0], 'face')
self.assertEqual(face[1], [(113, 40), (450, 40), (450, 403),
(113, 403)])
def loadPascalXMLByFilename(self, xmlPath):
if self.filename is None:
return
if os.path.isfile(xmlPath) is False:
return
tVocParseReader = PascalVocReader(xmlPath)
shapes = tVocParseReader.getShapes()
self.loadLabels(shapes)
def loadPascalXMLByFilename(self, xmlPath):
if self.filePath is None:
return
if os.path.isfile(xmlPath) is False:
return
tVocParseReader = PascalVocReader(xmlPath)
shapes = tVocParseReader.getShapes()
self.loadLabels(shapes)
def check_label(self):
flist = os.listdir(self.anno_path)
for filename in flist:
filepath = self.anno_path + '/' + filename
pr = PR(filepath)
shapes = pr.getShapes()
sl = pr.getSources()
if len(shapes) == 0:
print(filepath)
def convert(image_path, annotation_path):
save_path = os.path.join(os.path.dirname(annotation_path), "xml_converted")
if not os.path.exists(save_path): os.makedirs(save_path)
for file in os.listdir(annotation_path):
if file.endswith(".xml"):
annotation_no_xml = os.path.splitext(file)[0]
imagePath = image_path + "/" + annotation_no_xml + ".jpg"
image = QImage()
image.load(imagePath)
imageShape = [
image.height(),
image.width(), 1 if image.isGrayscale() else 3
]
imgFolderName = os.path.basename(image_path)
imgFileName = os.path.basename(imagePath)
writer = YOLOWriter(imgFolderName,
imgFileName,
imageShape,
localImgPath=imagePath)
# Read classes.txt
classListPath = annotation_path + "/" + "classes.txt"
classesFile = open(classListPath, 'r')
classes = classesFile.read().strip('\n').split('\n')
classesFile.close()
# Read VOC file
filePath = annotation_path + "/" + file
tVocParseReader = PascalVocReader(filePath)
shapes = tVocParseReader.getShapes()
num_of_box = len(shapes)
for i in range(num_of_box):
label = classes.index(shapes[i][0])
xmin = shapes[i][1][0][0]
ymin = shapes[i][1][0][1]
x_max = shapes[i][1][2][0]
y_max = shapes[i][1][2][1]
writer.addBndBox(xmin, ymin, x_max, y_max, label, 0)
writer.save(targetFile=save_path + "/" + annotation_no_xml +
".txt")
def split_single_image(self, img_id):
xml_file = self.anno_path / f"{img_id}.xml"
objects = PascalVocReader(xml_file).get_shapes()
img_file = self.img_path / f"{img_id}{self.suffix}"
# img = cv2.imread(str(img_file))
# img_height, img_width = img.shape[:2]
img = gdal.Open(str(img_file))
img_height = img.RasterYSize
img_width = img.RasterXSize
top, left = 0, 0
# start_positions = []
patch_ids = []
while (top < img_height):
# print(top)
if (top + self.patch_size >= img_height):
top = max(img_height - self.patch_size, 0)
left = 0
while (left < img_width):
# print(f"left = {left}")
if (left + self.patch_size >= img_width):
left = max(img_width - self.patch_size, 0)
right = min(left + self.patch_size, img_width)
down = min(top + self.patch_size, img_height)
patch_name = f"{img_id}__{top}__{left}"
self.save_patches(img, objects, patch_name, left, top, right,
down)
patch_ids.append(patch_name)
if left + self.patch_size >= img_width:
break
else:
left += self.slide_step
if top + self.patch_size >= img_height:
break
else:
top += self.slide_step
img = None
print(img_file)
if img_id in self.train_sets:
# self.train_ids += patch_ids
with (self.set_out_path / 'train.txt').open('a') as fp:
fp.write('\n'.join(patch_ids) + '\n')
else:
# self.val_ids += patch_ids
with (self.set_out_path / 'val.txt').open('a') as fp:
fp.write('\n'.join(patch_ids) + '\n')
def test_upper(self):
dir_name = os.path.abspath(os.path.dirname(__file__))
libs_path = os.path.join(dir_name, '..', 'libs')
sys.path.insert(0, libs_path)
from pascal_voc_io import PascalVocWriter
from pascal_voc_io import PascalVocReader
# Test Write/Read
writer = PascalVocWriter('tests', 'test', (512, 512, 1), localImgPath='tests/test.bmp')
difficult = 1
writer.addBndBox(60, 40, 430, 504, 'person', difficult)
writer.addBndBox(113, 40, 450, 403, 'face', difficult)
writer.save('tests/test.xml')
reader = PascalVocReader('tests/test.xml')
shapes = reader.getShapes()
personBndBox = shapes[0]
face = shapes[1]
self.assertEqual(personBndBox[0], 'person')
self.assertEqual(personBndBox[1], [(60, 40), (430, 40), (430, 504), (60, 504)])
self.assertEqual(face[0], 'face')
self.assertEqual(face[1], [(113, 40), (450, 40), (450, 403), (113, 403)])
imgFolderName = os.path.basename(imgFolderPath)
imgFileName = os.path.basename(imagePath)
writer = YOLOWriter(imgFolderName,
imgFileName,
imageShape,
localImgPath=imagePath)
# Read classes.txt
classListPath = imgFolderPath + "/" + "classes.txt"
classesFile = open(classListPath, 'r')
classes = classesFile.read().strip('\n').split('\n')
classesFile.close()
# Read VOC file
filePath = imgFolderPath + "/" + file
tVocParseReader = PascalVocReader(filePath)
shapes = tVocParseReader.getShapes()
num_of_box = len(shapes)
for i in range(num_of_box):
label = classes.index(shapes[i][0])
xmin = shapes[i][1][0][0]
ymin = shapes[i][1][0][1]
x_max = shapes[i][1][2][0]
y_max = shapes[i][1][2][1]
writer.addBndBox(xmin, ymin, x_max, y_max, label, 0)
writer.save(targetFile=imgFolderPath + "/" + annotation_no_xml +
".txt")
from unittest import TestCase
import sys
import os
dir_name = os.path.abspath(os.path.dirname(__file__))
libs_path = os.path.join(dir_name, '..', 'libs')
sys.path.insert(0, libs_path)
from pascal_voc_io import PascalVocWriter
from pascal_voc_io import PascalVocReader
# Test Write/Read
writer = PascalVocWriter('tests',
'test', (512, 512, 1),
localImgPath='tests/test.bmp')
difficult = 1
writer.addBndBox(60, 40, 430, 504, 'person', difficult)
writer.addBndBox(113, 40, 450, 403, 'face', difficult)
writer.save('tests/test.xml')
reader = PascalVocReader('tests/test.xml')
shapes = reader.getShapes()
from pascal_voc_io import PascalVocWriter
from pascal_voc_io import PascalVocReader
import os
import cv2
path = '/root/darknet/scripts/VOCdevkit/VOC2050back/Annotations/'
imgpath = '/root/darknet/scripts/VOCdevkit/VOC2050back/JPEGImages/'
pathdir = os.listdir(path)
for file in pathdir:
filename = path + file
print file
dot = file.find('.')
purename = file[:dot]
reader = PascalVocReader(filename)
reader.parseXML()
boxes = []
for i, shape in enumerate(reader.shapes):
if i < (len(reader.shapes) / 2):
boxes.append(shape)
boxes = sorted(boxes, key=lambda item: item[1][0][0])
readimgname = 'VOCdevkit/VOC2050/JPEGImages/{}.jpg'.format(purename)
img = cv2.imread(readimgname)
imgSize = [img.shape[0], img.shape[1], 1]
foldername = 'JPEGImages'
filename = purename + '.jpg'
localImagePath = '/root/darknet/scripts/VOCdevkit/VOC2050back/JPEGImages/' + filename
writer = PascalVocWriter(foldername,
filename,
imgSize,
localImgPath=localImagePath)
#print type(boxes)
L.append(os.path.join(file))
return L
file_dir = '/root/darknet/scripts/VOCdevkit/VOC2050_corrected/JPEGImages/'
goal_dir = '/root/darknet/scripts/VOCdevkit/VOC2050_corrected/JPEGImages/'
xml_dir = '/root/darknet/scripts/VOCdevkit/VOC2050_corrected/Annotations/'
foldername = 'JPEGImages'
for name in file_name(file_dir):
img = cv2.imread(file_dir + name)
dot = name.find('.')
namenum = int(name[:dot])
newname = str(namenum + 200000)
xmlfname = xml_dir + name[:dot] + '.xml'
print xmlfname
reader = PascalVocReader(xmlfname)
reader.parseXML()
boxes = []
for i, shape in enumerate(reader.shapes):
if i < (len(reader.shapes) / 2):
boxes.append(shape)
#print boxes
for box in boxes:
#print '--------'
#print box
section = random.randint(1, 4)
xmin = box[1][0][0]
ymin = box[1][0][1]
xmax = box[1][2][0]
im = 128 * np.ones((300, 500, 3), dtype=np.uint8)
for i in xrange(2):
_, _= im_detect(net, im)
im_names = os.listdir(input_dir+"/img")
for im_name in im_names:
print '~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~'
print "auto label {0}".format(im_name)
im_file = os.path.join(input_dir, 'img', im_name)
label_file = os.path.join(input_dir, 'annotation', im_name)
index = label_file.rindex('.')
label_file = label_file[:index] + ".xml"
if os.path.isfile(label_file):
tmp_reader = PascalVocReader(label_file)
size = tmp_reader.getSize()
tmp_writer = PascalVocWriter(input_dir, im_name, size)
shapes = tmp_reader.getShapes()
for i in shapes:
tmp_writer.addBndBox(i[1][0][0], i[1][0][1], i[1][2][0], i[1][2][1], i[0])
print "load {0} labels from {1}".format(len(shapes), label_file)
else:
print "can not find label file {0}".format(label_file)
size = cv2.imread(im_file).shape
tmp_writer = PascalVocWriter(input_dir, im_name, size)
n = auto_label(net, im_file, tmp_writer)
tmp_writer.save(output_dir + '/' + os.path.basename(label_file))
print "auto label {0} labels".format(n)