def segmentation(self, images, size, file_name,
collection): # collection_id
"""
检测分割
"""
ann_file = xml.dom.minidom.parseString(file_name) # 打开xml文档
annotation = ann_file.documentElement # 得到xml文件对象
obj_list = annotation.getElementsByTagName('object') # 获取object标签对象
segmented = annotation.getElementsByTagName('segmented')
obj_cnt = 0
ann_list = []
segmentation = segmented[0].firstChild.data
for idx in range(0, obj_list.length):
# segmentation = segmented[objCnt].firstChild.data
obj_cnt += 1
name = obj_list[idx].getElementsByTagName(
"name")[0].childNodes[0].data # 取出当前name标签对之间的数据
bndbox = obj_list[idx].getElementsByTagName("bndbox")[0]
xmin = bndbox.getElementsByTagName(
"xmin")[0].childNodes[0].data # 下左
xmax = bndbox.getElementsByTagName(
"xmax")[0].childNodes[0].data # 下右
ymin = bndbox.getElementsByTagName(
"ymin")[0].childNodes[0].data # 上左
ymax = bndbox.getElementsByTagName(
"ymax")[0].childNodes[0].data # 上右
labels = Label.query.filter_by(collection_id=collection.id,
name=name).first()
if not labels:
continue
# write anns in annStr
ann_str = '{'
ann_str += '"bbox": [' + xmin + ", " + ymin + ", " + \
str(int(xmax) - int(xmin)) + ', ' + str(int(ymax)-int(ymin)) + '], '
ann_str += '"category_id": ' + str(
labels.label_id
) + ',' + '"category_name": ' + '"' + name + '"' + ','
ann_str += '"segmentation": [' + "[" + str(
segmentation) + "]" + "]" + "},"
ann_list.append(ann_str)
next_join = '"classification": ' + '[],'
next_join += '"annotation": [' + ''.join(ann_list)[:-1] + ']}'
str_list = [self.join_images, next_join]
data = ''.join(str_list)
# '图片上传时间','xml文件'，宽高，图片id
up_file = data % (images.create_time, images.site[10:],
size.get('height'), size.get('width'), images.id)
file_json = fun.upload(up_file, file_ext_name='json')
images.status = 3
images.label_path = file_json.get('file_id')
db.session.add(images)
db.session.commit()
def default(self, images, size):
"""
未标记
"""
next_join = '"classification": [],'
next_join += '"annotation": []' + '}'
str_list = [self.join_images, next_join]
data = ''.join(str_list)
up_file = data % (images.create_time, images.site[10:],
size.get('height'), size.get('width'), images.id)
file_json = fun.upload(up_file, file_ext_name='json')
images.status = 0
images.label_path = file_json.get('file_id')
db.session.add(images)
db.session.commit()
def spider(
spider_dict
): # spider_dict 是后端从前端post表单获取到用户给的参数形式为{'keyword':xx, 'spider_page_num':xx, 'start_page':xx， 'number'：xx}
"""爬虫任务"""
from DataSet.rabbitMQ.task_queue_client import spider_task
with app.app_context():
site_list = list()
print spider_dict
spider_image_list = spider_task(spider_dict)
spider_image_list = json.loads(spider_image_list)
print spider_image_list
for spider_image in spider_image_list:
file = urllib2.urlopen(spider_image)
tmpIm = cStringIO.StringIO(file.read())
# c = tmpIm.read()s
site = fun.upload(tmpIm.read(), file_ext_name='jpg')
site_list.append(site)
# 数据库存储
for spider_file in site_list:
image = Image()
image.name = spider_file.get('filename')
image.site = spider_file.get('file_id')
spider_dict1 = json.loads(spider_dict)
print(spider_dict1['collection_id'])
image.collection_id = spider_dict1['collection_id']
print image.collection_id
db.session.add(image)
db.session.commit()
images = Image.query.filter_by(name=image.name).first()
image_url_path = current_app.config[
'NGINX_SITE'] + fun.getInfo(images.site)['group_file_id']
file = urllib2.urlopen(image_url_path)
size = image_size(file.read())
clf.default(images, size)
collection_id = spider_dict1['collection_id']
same_images_clean(collection_id)
return site_list
def classification(self, images, size, category_id, category_name):
"""
分类/人脸识别
"""
next_join = '"classification": [' + '{'
next_join += '"category_id": %s,' + '"category_name": "%s"' + '}' + '],'
next_join += '"annotation": []' + '}'
strlist = [self.join_images, next_join]
data = ''.join(strlist)
up_file = data % (images.create_time, images.site[10:],
size.get('height'), size.get('width'), images.id,
category_id, category_name)
print 111
print type(up_file)
print up_file
file_json = fun.upload(up_file, file_ext_name='json')
print 222
print file_json
images.status = 3
images.label_path = file_json.get('file_id')
db.session.add(images)
db.session.commit()
def preprocessing(collection_id, label_info):
"""预标注任务"""
from DataSet.rabbitMQ.task_queue_client import preprocessing_task
with app.app_context(): # 开辟一个上下文管理空间
# get config
model = -1
pre_model = list()
label_name_list = list()
c_label_name_list = list()
for key in label_info:
model = int(key.split('_')[0])
pre_model.append(int(key.split('_')[1]))
for k in label_info[key]:
label_name_list.append(label_info[key][k])
c_label_name_list.append(k)
images = None
try:
images = Image.query.filter_by(collection_id=collection_id,
status=1).all()
except Exception as e:
current_app.logger.error(e)
if not images:
print(label_info, 'no images')
for image in images:
# get image url
image_obj = Image.query.filter_by(id=image.id).first()
image_url_path = current_app.config['NGINX_SITE'] + fun.getInfo(
image_obj.site)['group_file_id']
file = urllib2.urlopen(image_url_path) # 获取图片
# get image file
tmpIm = cStringIO.StringIO(file.read(
)) # Return a StringIO-like stream for reading or writing
img = image_pil.open(tmpIm) # PIL模块处理图像
if img.mode in ('RGBA', 'LA', 'P'):
background = image_pil.new('RGB', img.size)
background.paste(img)
img = background
# base64
bs = BytesIO()
img.save(bs, format="JPEG")
b64 = base64.b64encode(bs.getvalue())
# get request
type_b64_list = list()
type_b64_list.append(b64)
type_b64_list.append(model)
type_b64_list.append(pre_model)
# request
json_data = preprocessing_task(type_b64_list) # 指定预处理队列处理
json_data = json.loads(json_data)
# print( 'label results ', json_data )
# get label
label_url_path = current_app.config['NGINX_SITE'] + fun.getInfo(
image_obj.label_path)['group_file_id']
# 读取json文件内容
file = urllib2.urlopen(label_url_path)
tmpIm = cStringIO.StringIO(file.read())
label_data = tmpIm.read()
label_json_data = json.loads(label_data)
# print( 'old label_json_data ', label_json_data )
isSuccess = False
# 模型1
if model == 1:
labels = list()
for key in json_data:
labels = labels + json_data[key]['labels']
for label_dict in labels:
if label_dict['content'] in c_label_name_list:
label_name = label_name_list[c_label_name_list.index(
label_dict['content'])]
label = Label.query.filter_by(
collection_id=collection_id,
name=label_name).first()
# 创建要保存的标签字典
classification = dict()
classification['category_id'] = label.label_id
classification['category_name'] = label_name
label_json_data['classification'] = classification
label.count += 1
isSuccess = True
else:
print('other label: ', label_dict['content'])
# 模型2
elif model == 2:
detection = list()
for key in json_data:
detection = detection + json_data[key]['detection']
for label_dict in detection:
pmid = label_dict['pre_model']
if float(label_dict['score']) > 0.3:
AI_label_name = AI_label_list[pmid][
label_dict['label_id']].decode('utf-8')
# print( 'c_label_name_list', AI_label_name, c_label_name_list )
if AI_label_name in c_label_name_list:
label_name = label_name_list[
c_label_name_list.index(AI_label_name)]
label = Label.query.filter_by(
collection_id=collection_id,
name=label_name).first()
# 标注框坐标转换
bbox_list = list()
bbox_list.append('%.2f' %
(float(label_dict['xmin'])))
bbox_list.append('%.2f' %
(float(label_dict['ymin'])))
bbox_list.append('%.2f' %
(float(label_dict['xmax']) -
float(label_dict['xmin'])))
bbox_list.append('%.2f' %
(float(label_dict['ymax']) -
float(label_dict['ymin'])))
# 创建json annotation字典
annotation = dict()
annotation['bbox'] = bbox_list
annotation['category_id'] = label.label_id
annotation['category_name'] = label.name
annotation['segmentation'] = []
if 'feaData' in label_dict:
annotation['feaData'] = label_dict['feaData']
label_json_data['annotation'].append(annotation)
label.count += 1
isSuccess = True
if isSuccess:
# print( 'label_json_data', label_json_data )
label_data = json.dumps(label_json_data)
# 保存修改后的json文件
new_label_path = fun.upload(label_data, file_ext_name='json')
# 删除原json文件
fun.remove(image.label_path)
# 修改数据库json文件存储路径
image_obj.label_path = new_label_path['filename']
# 修改图片状态
image_obj.status = 2
db.session.commit()
else:
image_obj.status = 2
db.session.commit()
def save_label():
"""标注保存"""
data = request.get_json()
state = data.get('classification')
collection_id = data.get('collection_id')
collection_type = Collection.query.filter_by(id=collection_id).first().type
# 分类/人脸识别标注
if state:
label_id = state['category_id']
category_name = state['category_name']
if category_name is not None:
try:
label = Label.query.filter_by(collection_id=collection_id,
label_id=label_id).first()
except Exception as e:
current_app.logger.error(e)
return jsonify(err_no=RET.DBERR, err_desc='查询失败')
if not label:
return jsonify(err_no=RET.NODATA, err_desc='没有该标签')
# 创建要保存的标签字典
classification = dict()
classification['category_id'] = label_id
classification['category_name'] = category_name
image_site = data.get('image')
try:
image = Image.query.filter_by(
site=str(image_site).split('/', 4)[4]).first()
except:
current_app.logger.error(image_site)
return jsonify(err_no=RET.DBERR, err_desc='查询失败')
if image.status == 3:
return jsonify(err_no=RET.NODATA, err_desc='请勿重复标记分类图片')
label_url_path = current_app.config['NGINX_SITE'] + fun.getInfo(
image.label_path)['group_file_id']
# 读取json文件内容
file = urllib2.urlopen(label_url_path)
tmpIm = cStringIO.StringIO(file.read())
label_date = tmpIm.read()
label_json_date = json.loads(label_date)
label_json_date['classification'] = [classification]
label_date = json.dumps(label_json_date)
# 保存修改后的json文件
new_label_path = fun.upload(label_date, file_ext_name='json')
# 删除原json文件
fun.remove(image.label_path)
# 修改数据库json文件存储路径
image.label_path = new_label_path['filename']
# 修改图片状态
if collection_type == 1:
pass
# image.status = 3
label.count += 1
db.session.commit()
else:
return jsonify(
err_no=RET.PARAMERR,
err_desc='参数缺失',
)
# 检测/分割标注
state = data.get('annotation')
if state:
for label_dict in data.get('annotation'):
category_id = label_dict['category_id']
category_name = label_dict['category_name']
bbox = label_dict['bbox']
segmentation = label_dict['segmentation']
if category_name and bbox is not None:
try:
label = Label.query.filter_by(
collection_id=collection_id,
label_id=category_id).first()
except Exception as e:
current_app.logger.error(e)
return jsonify(err_no=RET.DBERR, err_desc='查询失败')
if not label:
return jsonify(err_no=RET.NODATA, err_desc='没有该标签')
try:
for num in bbox:
num = float(num)
if math.isnan(num):
current_app.logger.error(num)
return jsonify(err_no=RET.DBERR, err_desc='参数错误')
except Exception as e:
current_app.logger.error(e)
return jsonify(err_no=RET.DBERR, err_desc='参数错误')
# 创建要保存的标签字典
annotation = dict()
annotation['bbox'] = bbox
annotation['category_id'] = category_id
annotation['category_name'] = category_name
annotation['segmentation'] = segmentation
image_site = data.get('image')
image = Image.query.filter_by(
site=str(image_site).split('/', 4)[4]).first()
label_url_path = current_app.config['NGINX_SITE'] + fun.getInfo(
image.label_path)['group_file_id']
# 读取json文件内容
file = urllib2.urlopen(label_url_path)
tmpIm = cStringIO.StringIO(file.read())
label_date = tmpIm.read()
label_json_date = json.loads(label_date)
label_json_date['annotation'].append(annotation)
label_date = json.dumps(label_json_date)
# 保存修改后的json文件
new_label_path = fun.upload(label_date, file_ext_name='json')
# 删除原json文件
fun.remove(image.label_path)
# 修改数据库json文件存储路径
image.label_path = new_label_path['filename']
# 修改图片状态
if collection_type == 2:
pass
# image.status = 3
label.count += 1
db.session.commit()
else:
return jsonify(err_no=RET.PARAMERR, err_desc='参数缺失')
return jsonify(err_no=RET.OK, err_desc='标注成功')
def delete_label():
"""删除标签"""
data = request.get_json()
state = data.get('classification')
collection_id = data.get('collection_id')
if state:
label_id = state['category_id']
try:
label = Label.query.filter_by(collection_id=collection_id,
label_id=label_id).first()
except Exception as e:
current_app.logger.error(e)
return jsonify(err_no=RET.DBERR, err_desc='查询失败')
if not label:
return jsonify(err_no=RET.NODATA, err_desc='没有该标签')
image_site = data.get('image')
image = Image.query.filter_by(
site=str(image_site).split('/', 4)[4]).first()
label_url_path = current_app.config['NGINX_SITE'] + fun.getInfo(
image.label_path)['group_file_id']
# 读取json文件内容
file = urllib2.urlopen(label_url_path)
tmpIm = cStringIO.StringIO(file.read())
label_date = tmpIm.read()
label_json_date = json.loads(label_date)
old_label_name = label_json_date['classification'][0]['category_name']
old_label = Label.query.filter_by(collection_id=collection_id,
name=old_label_name).first()
old_label.count -= 1
del label_json_date['classification'][0]
label_date = json.dumps(label_json_date)
# 保存修改后的json文件
new_label_path = fun.upload(label_date, file_ext_name='json')
# 删除原json文件
fun.remove(image.label_path)
image.label_path = new_label_path['filename']
db.session.commit()
# 检测/分割标注
for label_dict in data.get('annotation'):
category_id = label_dict['category_id']
print('label_dict', label_dict)
old_bbox = label_dict['old_bbox']
old_bbox = [float(d) for d in old_bbox.split(',')]
try:
label = Label.query.filter_by(collection_id=collection_id,
label_id=category_id).first()
except Exception as e:
current_app.logger.error(e)
return jsonify(err_no=RET.DBERR, err_desc='查询失败')
if not label:
return jsonify(err_no=RET.NODATA, err_desc='没有该标签')
image_site = data.get('image')
image = Image.query.filter_by(
site=str(image_site).split('/', 4)[4]).first()
label_url_path = current_app.config['NGINX_SITE'] + fun.getInfo(
image.label_path)['group_file_id']
# 读取json文件内容
file = urllib2.urlopen(label_url_path)
tmpIm = cStringIO.StringIO(file.read())
label_date = tmpIm.read()
label_json_date = json.loads(label_date)
for label_json_dict in label_json_date['annotation']:
bbox = [float(d) for d in label_json_dict['bbox']]
if bbox == old_bbox:
print('old_bbox = ', old_bbox)
old_label_id = label_dict['category_id']
old_label = Label.query.filter_by(
collection_id=collection_id,
label_id=old_label_id).first()
old_label.count -= 1
label_json_date['annotation'].remove(label_json_dict)
label_date = json.dumps(label_json_date)
# 保存修改后的json文件
new_label_path = fun.upload(label_date, file_ext_name='json')
# 删除原json文件
fun.remove(image.label_path)
# 修改数据库json文件存储路径
image.label_path = new_label_path['filename']
db.session.commit()
return jsonify(err_no=RET.OK, err_desc='删除成功')
def change_label():
"""改变标签"""
data = request.get_json()
state = data.get('classification')
collection_id = data.get('collection_id')
if state:
label_id = state['category_id']
category_name = state['category_name']
if category_name is not None:
try:
label = Label.query.filter_by(collection_id=collection_id,
label_id=label_id).first()
except Exception as e:
current_app.logger.error(e)
return jsonify(err_no=RET.DBERR, err_desc='查询失败')
if not label:
return jsonify(err_no=RET.NODATA, err_desc='没有该标签')
classification = dict()
classification['category_id'] = label_id
classification['category_name'] = category_name
image_site = data.get('image')
image = Image.query.filter_by(
site=str(image_site).split('/', 4)[4]).first()
label_url_path = current_app.config['NGINX_SITE'] + fun.getInfo(
image.label_path)['group_file_id']
# 读取json文件内容
file = urllib2.urlopen(label_url_path)
tmpIm = cStringIO.StringIO(file.read())
label_date = tmpIm.read()
label_json_date = json.loads(label_date)
old_label_name = label_json_date['classification'][0][
'category_name']
old_label = Label.query.filter_by(collection_id=collection_id,
name=old_label_name).first()
old_label.count -= 1
del label_json_date['classification'][0]
label_json_date['classification'].append(classification)
label_date = json.dumps(label_json_date)
# 保存修改后的json文件
new_label_path = fun.upload(label_date, file_ext_name='json')
# 删除原json文件
fun.remove(image.label_path)
image.label_path = new_label_path['filename']
label.count += 1
db.session.commit()
else:
return jsonify(err_no=RET.PARAMERR, err_desc='参数缺失')
# 检测/分割标注
for label_dict in data.get('annotation'):
category_id = label_dict['category_id']
category_name = label_dict['category_name']
print(label_dict)
bbox = label_dict['bbox']
old_bbox = label_dict['old_bbox']
segmentation = label_dict['segmentation']
if category_name and bbox and segmentation and old_bbox is not None:
try:
label = Label.query.filter_by(collection_id=collection_id,
label_id=category_id).first()
except Exception as e:
current_app.logger.error(e)
return jsonify(err_no=RET.DBERR, err_desc='查询失败')
if not label:
return jsonify(err_no=RET.NODATA, err_desc='没有该标签')
try:
for num in bbox:
if not num.isnumeric():
current_app.logger.error(str(num))
return jsonify(err_no=RET.DBERR, err_desc='参数错误')
num = float(num)
except Exception as e:
current_app.logger.error(e)
return jsonify(err_no=RET.DBERR, err_desc='参数错误')
# 创建要保存的标签字典
annotation = dict()
annotation['bbox'] = bbox
annotation['category_id'] = category_id
annotation['category_name'] = category_name
annotation['segmentation'] = segmentation
image_site = data.get('image')
image = Image.query.filter_by(
site=str(image_site).split('/', 4)[4]).first()
label_url_path = current_app.config['NGINX_SITE'] + fun.getInfo(
image.label_path)['group_file_id']
# 读取json文件内容
file = urllib2.urlopen(label_url_path)
tmpIm = cStringIO.StringIO(file.read())
label_date = tmpIm.read()
label_json_date = json.loads(label_date)
for label_json_dict in label_json_date['annotation']:
if label_json_dict['bbox'] == old_bbox:
old_label_name = label_dict['category_name']
old_label = Label.query.filter_by(
collection_id=collection_id,
old_label_name=old_label_name).first()
old_label.count -= 1
label_json_date['annotation'].remove(label_json_dict)
label_json_date['annotation'].append(annotation)
label_date = json.dumps(label_json_date)
# 保存修改后的json文件
new_label_path = fun.upload(label_date, file_ext_name='json')
# 删除原json文件
fun.remove(image.label_path)
# 修改数据库json文件存储路径
image.label_path = new_label_path['filename']
label.count += 1
db.session.commit()
else:
return jsonify(RET.PARAMERR, err_desc='参数缺失')
return jsonify(err_no=RET.OK, err_desc='修改成功')