def add_box(path,boxes):
# print(path)
tree = ET.ElementTree(file=path)
root = tree.getroot()
for bndbox in boxes:
root=XML.add_tag(root,bndbox)
XML.write_xml(tree,path)
def calculate_similarity_on_labels(baseName, checkName):
# 1. Match labels
global labelNumber
scoreSoles = 0
baseXml = os.path.join(xmlDir, baseName)
checkXml = os.path.join(xmlDir, checkName)
baseObjs = XML.read_objects(baseXml)
checkObjs = XML.read_objects(checkXml)
for checkObj in checkObjs:
if checkObj['name'] in forceFetch:
return 0
labelNumber += len(baseObjs) + len(checkObjs)
pairs, baseSoles, checkSoles = match_labels(baseObjs, checkObjs)
# 2. Calculate similarity of each pair labels which containing roi similarity and location similarity
availablePairLenth = len(pairs)
# for pair in pairs:
# if pair[0]['name'] in ignoreLabels:
# availablePairLenth-=1
if availablePairLenth > 0:
similarity = pair_similarity(pairs, baseName, checkName)
for baseSole in baseSoles:
scoreSoles += 1 / weight[baseSole['name']]
for checkSole in checkSoles:
scoreSoles += 1 / weight[checkSole['name']]
print('mother={}+{}+{}'.format(availablePairLenth, scoreSoles,
scoreSoles))
similarity = similarity / (availablePairLenth + scoreSoles +
scoreSoles)
return similarity
else:
# print(baseSole)
# print(checkSole)
return 0
def regular_xml(path, filename):
global count
tree = ET.ElementTree(file=path)
root = tree.getroot()
XML.chk_label(root, sizeThreshDict)
XML.chag_size(root, filename, width, height)
XML.chag_filename(root, filename)
XML.thresh_size(root, width, sizeThreshDict)
XML.write_xml(tree, path)
def regular_xml(path, filename):
global count
tree = ET.ElementTree(file=path)
root = tree.getroot()
# XML.chag_label(root,labelDict) # 修改label名称
# root=XML.del_tag(root,names) # 删除指定label
root = XML.save_tag(root, names) # 只保留指定label
# root=XML.thresh_size(path,root,sizeThresh,sizeThreshNames)
XML.write_xml(tree, path)
def add_new_cates(xmlPath,referPath,addCates):
addBoxes=XML.read_objects(referPath)
tree = ET.ElementTree(file=xmlPath)
root = tree.getroot()
root=XML.del_tag(root,addCates)
# print(addBoxes)
for box in addBoxes:
if box['name'] in addCates:
root=XML.add_tag(root,box)
XML.write_xml(tree,xmlPath)
def calculate_similarity_on_labels(baseName, checkName):
# 1. Match labels
global labelNumber
scoreSoles = 0
baseXml = os.path.join(xmlDir, baseName)
checkXml = os.path.join(xmlDir, checkName)
baseObjs = XML.read_objects_exclude_minsize(baseXml)
checkObjs = XML.read_objects_exclude_minsize(checkXml)
for checkObj in checkObjs:
if checkObj['name'] in forceFetch:
return 100
labelNumber += len(baseObjs) + len(checkObjs)
similarity = match_labels(baseObjs, checkObjs)
return similarity
def regular_xml(newxmlPath, img, newimg):
print(newxmlPath, xml, newxml)
tree = ET.ElementTree(file=newxmlPath)
root = tree.getroot()
for obj in root.findall('filename'):
print(obj.text)
obj.text = newimg
for obj in root.findall('path'):
print(obj.text)
obj.text = obj.text.replace(img, newimg)
XML.write_xml(tree, newxmlPath)
def filter_by_cate(allXmls, cateName):
filteredXmls = []
viewCount = 0
xmlSum = len(allXmls)
for xml in allXmls:
xmlPath = os.path.join(xmlDir, xml)
objs = XML.read_objects(xmlPath)
for obj in objs:
objName = obj['name']
if objName == cateName:
filteredXmls.append(xml)
viewCount += 1
sys.stdout.write('\rFiltering xmls >>{count:.2f}%'.format(
count=100 * float(viewCount) / xmlSum))
# sys.stdout.write('\r>> Detecting {name} {count:.2f}%'.format(name=videoPath.split('/')[-1],count=(100*float(i)/total_frames)))
sys.stdout.flush()
return filteredXmls
def analyze_xml(path):
objs = XML.read_objects(path)
#print(objs)
for obj in objs:
name = obj['name']
'''
if name in ['、', 'microbus_foreign\n', 'micorbus_foreign', 'car_pickup', 'truck_foregin', 'suv_foreign']:
print(path)
print(obj)
exit()
'''
if name not in count.info:
count.addtag(name)
else:
count.update(name)
count.count()
xmin = int(obj['xmin'])
ymin = int(obj['ymin'])
xmax = int(obj['xmax'])
ymax = int(obj['ymax'])
w = xmax - xmin
h = ymax - ymin
scale = round(np.sqrt(w * h))
nn = int(scale / 25) + 1
scale = nn * 25
'''
if scale >2000:
print(path)
print(obj)
exit()
'''
if scale not in count2.info:
count2.addtag(scale)
else:
count2.update(scale)
count2.count()
augNameList=['close','cover','open']
xmlDir=r"/DATACENTER2/yh/detectron/caffe2/train/safebox/faster_voc/11469/xml_1713"
tarDir=r"/DATACENTER2/yh/detectron/caffe2/train/safebox/faster_voc/11469/xml_1713_aug"
FILES.mkdir(tarDir)
allXmls=[x for x in FILES.get_sorted_files(xmlDir) if ".xml" in x]
cur=0
bgn=800 #包含
end=1900 #包含
for xml in allXmls:
cur+=1
# print(cur) #第cur个文件
if cur>=bgn and cur<=end:
xmlPath=xmlDir+'/'+xml
tarXmlPath=tarDir+'/'+xml
shutil.copy(xmlPath,tarXmlPath)
BBOXES=XML.read_object(xmlPath)
for augRate_xmin in range(0,5):
augValue_xmin=augRate_xmin*0.02
for augRate_xmax in range(0,5):
augValue_xmax=augRate_xmax*0.02
for augRate_ymin in range(0,5):
augValue_ymin=augRate_ymin*0.02
for augRate_ymax in range(0,5):
augValue_ymax=augRate_ymax*0.02
if augRate_xmin+augRate_xmax+augRate_ymin+augRate_ymax!=0:
bboxes=[]
for i in range(len(BBOXES)):
if BBOXES[i][0] in augNameList and check_overlap(BBOXES[i],BBOXES):
bboxes.append(trans_box(BBOXES[i],augValue_xmin,augValue_ymin,augValue_xmax,augValue_ymax))
try:
add_box(tarXmlPath,bboxes)
if top_labels:
#outImg
image, labels, xmins, ymins, xmaxs, ymaxs = drawImg(
image, top_labels, top_xmin, top_ymin, top_xmax, top_ymax,
top_scores)
# desImg = os.path.join(outImg,file)
# cv2.imwrite(desImg,image)
# # outXml
if len(labels) > 0:
tarXml = os.path.join(tarXmlDir, file.split('.')[0] + '.xml')
shutil.copy(referXml, tarXml)
tree = ET.ElementTree(file=tarXml)
root = tree.getroot()
for i in range(len(labels)):
# if labels[i] not in ["close","open","cover"]:
if 1 > 0:
xmin = str(xmins[i])
ymin = str(ymins[i])
xmax = str(xmaxs[i])
ymax = str(ymaxs[i])
# print(labels[i],xmin,ymin,xmax,ymax)
obj = {
'name': labels[i],
'xmin': xmin,
'ymin': ymin,
'xmax': xmax,
'ymax': ymax
}
XML.add_tag(root, obj)
XML.write_xml(tree, tarXml)
def changeXmlLabelName(path, labelDict):
tree = ET.ElementTree(file=path)
root = tree.getroot()
root = XML.del_tag(root, ["13"])
root = XML.chag_name(root, labelDict)
XML.write_xml(tree, path)
def countObj(path):
boxes = XML.read_object(path)
return (len(boxes))
def regular_xml(path, changeLabelDict):
global count
tree = ET.ElementTree(file=path)
root = tree.getroot()
XML.chag_label(root, changeLabelDict, tag="object")
XML.write_xml(tree, path)
labelXmlDir = r'/disk2/hao.yang/project/Qin/data/imgs/isle/modelxml_isle_17117_havered/'
# checkedDir='/disk2/hao.yang/project/Qin/data/xmls/checkout/complete'
attentionNames = ['red']
confuseNames = ['yellow', 'other', 'blue']
count = 0
allChecked = []
# allChecked=[x for x in FILES.get_sorted_files(checkedDir) if ".xml" in x]
allXmls = [
x for x in FILES.get_sorted_files(modelXmlDir)
if ".xml" in x and x not in allChecked
]
for xml in allXmls:
labelXmlPath = os.path.join(labelXmlDir, xml)
modelXmlPath = os.path.join(modelXmlDir, xml)
if os.path.exists(labelXmlPath):
labelObjs = XML.read_objects(labelXmlPath)
modelObjs = XML.read_objects(modelXmlPath)
newObjs = []
for labelObj in labelObjs:
if labelObj['name'] in confuseNames:
newObjs.append(fuse(labelObj, modelObjs))
else:
newObjs.append(labelObj)
tree = ET.ElementTree(file=labelXmlPath)
root = tree.getroot()
XML.del_all_tag(root)
for box in newObjs:
root = XML.add_tag(root, box)
XML.write_xml(tree, labelXmlPath)
# FILES.copy_files_refer_dir(imgDir,'.jpg',badDir,badDir)
# font_label = ImageFont.truetype('/DATACENTER2/yh/resources/heiti.TTF',50)
# img_PIL = Image.fromarray(cv2.cvtColor(image,cv2.COLOR_BGR2RGB)) # cv2==>PIL
# draw = ImageDraw.Draw(img_PIL)
# draw.text((xmin,ymax-50),label_name,font_color,font=font_label)
# image = cv2.cvtColor(numpy.asarray(img_PIL),cv2.COLOR_RGB2BGR) # PIL==>cv2
# img_PIL = Image.fromarray(cv2.cvtColor(image,cv2.COLOR_BGR2RGB)) # cv2==>PIL
# draw = ImageDraw.Draw(img_PIL)
# image = cv2.cvtColor(numpy.asarray(img_PIL),cv2.COLOR_RGB2BGR) # PIL==>cv2
return image
if __name__ == "__main__":
start = time.time()
jpgDir = r"/disk2/hao.yang/project/Qin/data/xmls/isle/complete/imgs_overlook_16000_test/"
xmlDir = r"/disk2/hao.yang/project/Qin/data/xmls/isle/complete/isle_FMXX_test_16000/"
outDir = r"/disk2/hao.yang/project/Qin/data/xmls/isle/complete/imgs_overlook_16000_test_truth/"
FILES.rm_mkdir(outDir)
allJpgs = FILES.get_files(jpgDir)
for jpg in allJpgs:
# print(jpg)
if '.jpg' in jpg:
jpgPath = os.path.join(jpgDir, jpg)
xmlPath = os.path.join(xmlDir, jpg.split('.')[0] + '.xml')
objs = XML.read_objects(xmlPath)
image = cv2.imread(jpgPath)
img = drawXml(image, objs)
desImg = os.path.join(outDir, jpg)
cv2.imwrite(desImg, img)
# extraFrames(davDir,mp4Dir)
def checkName(path, name):
boxes = XML.read_object(path)
for box in boxes:
if box[0] == name:
return 1
return 0
bigNames = []
# bigNames不可能有漏标，如果漏掉，必然是故意的，所以不算错误
count = 0
allChecked = [x for x in FILES.get_sorted_files(checkedDir) if ".xml" in x]
if allChecked:
allXmls = [
x for x in FILES.get_sorted_files(labelXmlDir)
if ".xml" in x and x not in allChecked
]
else:
allXmls = [x for x in FILES.get_sorted_files(modelXmlDir) if ".xml" in x]
for xml in tqdm(allXmls):
labelXmlPath = os.path.join(labelXmlDir, xml)
modelXmlPath = os.path.join(modelXmlDir, xml)
if os.path.exists(labelXmlPath):
modelObjs = XML.read_objects(modelXmlPath)
labelObjs = XML.read_objects(labelXmlPath)
differ = 0
differ, newObjs = combine(modelObjs, labelObjs)
if differ:
badXmlPath = os.path.join(badDir, xml)
shutil.copy(labelXmlPath, badXmlPath)
# tree = ET.ElementTree(file=badXmlPath)
# root = tree.getroot()
# root=XML.del_all_tag(root)
# for boj in newObjs:
# root=XML.add_tag(root,boj)
# XML.write_xml(tree,badXmlPath)
count += differ
print(count)
# for modelObj in modelObjs:
