def ocr_result_visualization(_image_info, _box_info_list, _text_list):
"""
将检测的结果和识别的结果合并到一张图中
Args:
_image_info: 图片信息
_box_info_list: 所有box的列表
_text_list: 跟box顺序一致的识别结果
Returns: 合并后的图片的oss的路径
"""
to_return_result = {'bucket_name': '', 'path': ''}
oss_handler = get_oss_handler()
img = oss_handler.download_image_file(_image_info['bucket_name'],
_image_info['path'])
result_image = annotate_detect_rotated_bbox_and_text_result(
img, _box_info_list, _text_list, (0, 0, 255), 3)
date_string = get_date_string()
uuid_name = get_uuid_name()
image_path = os.path.join(date_string, uuid_name)
final_image_path = oss_handler.upload_image_file('result', image_path,
result_image, True, 50)
to_return_result['bucket_name'] = 'result'
to_return_result['path'] = final_image_path
return to_return_result
def face_parsing(_image_info, _face_box_info, _face_landmark_info):
"""
人脸语义分区
Args:
_image_info: 待识别的完整图像
_face_box_info: 人脸所在区域
_face_landmark_info: 人脸landmark坐标信息
Returns: 人脸不同区域的mask的key
"""
to_return_result = {
'parsing_info': {
'bucket_name': '',
'path': ''
},
}
oss_handler = get_oss_handler()
img = oss_handler.download_image_file(_image_info['bucket_name'],
_image_info['path'])
cropped_image = get_rotated_box_roi_from_image(img,
_face_box_info,
_scale_ratio=1.5)
face_parsing_result = face_parsing_handler.execute(cropped_image,
_face_landmark_info)
parsing_result = face_parsing_result['semantic_segmentation']
date_string = get_date_string()
name_string = get_uuid_name()
target_path = os.path.join(date_string, name_string)
target_path = oss_handler.upload_numpy_array('intermediate', target_path,
parsing_result)
to_return_result['parsing_info']['bucket_name'] = 'intermediate'
to_return_result['parsing_info']['path'] = target_path
return to_return_result
def execute(
self,
_video_url,
_extract_mode,
_interval_count,
_oss_helper: CloudObjectStorage,
_target_bucket=None,
):
"""
进行视频关键帧提取，并存在在oss中
Args:
_video_url: 视频地址
_extract_mode: 提取模式
_interval_count: 提取间隔
_oss_helper: oss
_target_bucket: 目标bucket name
Returns: 每个关键帧的bucket name和path
"""
cap = cv2.VideoCapture(_video_url)
video_fps = cap.get(cv2.CAP_PROP_FPS)
if _extract_mode == 'n_seconds':
interval_frames = int(video_fps * _interval_count)
elif _extract_mode == 'n_frames':
interval_frames = _interval_count
else:
raise VideoExtractMethodNotSupportException(
f'{_extract_mode} not support now')
video_frame_position = 0
date_string = get_date_string()
uuid_name = get_uuid_name()
all_tasks = []
with ThreadPoolExecutor() as executor:
while True:
grabbed, m_frame = cap.read()
if grabbed:
if _oss_helper and _target_bucket:
m_frame_target_bucket = _target_bucket
m_frame_target_path = os.path.join(
date_string, uuid_name, f'{video_frame_position}')
all_tasks.append(
executor.submit(_oss_helper.upload_image_file,
m_frame_target_bucket,
m_frame_target_path, m_frame,
False))
video_frame_position += interval_frames
cap.set(cv2.CAP_PROP_POS_FRAMES, video_frame_position)
else:
break
cap.release()
# 严格保证按顺序返回任务
for m_task in all_tasks:
if m_task.done():
# 这里原本设计是可以返回frame信息的，但考虑到整体架构，这样反而会降低程序的并行度，
# 提升单节点的耗时，所以将frame的返回直接去掉了。
yield _target_bucket, m_task.result()
def execute(
self, _to_download_url, _oss_helper: CloudObjectStorage = None,
_image_size_threshold=10,
):
"""
下载指定url的图像文件
Args:
_to_download_url: 待下载的图像url
_oss_helper: oss helper用于存储下载好的数据，可以为空
_image_size_threshold: 图像字节数(KB)限制，如果低于阈值会异常
Returns: 下载完成的结果
"""
to_return_result = OrderedDict()
try:
download_result_io = self.download_url(_to_download_url, 1024 * 2)
request_image = Image.open(download_result_io)
if request_image is None:
raise ImageFormatNotSupportException(
f'image:{_to_download_url} format not support,cannot decode by PILLOW')
image_c = len(request_image.getbands())
image_h = request_image.height
image_w = request_image.width
request_image_np = convert_pil_to_numpy(request_image)
if _image_size_threshold is not None and request_image_np.nbytes < _image_size_threshold * 1024:
raise ImageFileSizeAbnormalException(
f'image:{_to_download_url} is small than threshold,it may not be a normal picture')
# 有些情况是不需要存储的，可能直接就用了。
if _oss_helper:
file_name = get_uuid_name()
oss_path = os.path.join(get_date_string(), file_name)
# 存储原始图像
saved_path = _oss_helper.upload_image_file(self.bucket_name, oss_path, request_image,
_enable_compress=False)
else:
saved_path = ''
to_return_result['bucket_name'] = self.bucket_name
to_return_result['saved_path'] = saved_path
to_return_result['image_height'] = image_h
to_return_result['image_width'] = image_w
to_return_result['image_channel'] = image_c
to_return_result['image'] = request_image_np
return to_return_result
except requests.exceptions.ConnectionError as connect_error:
raise DownloadURLNotAvailableException(f'{_to_download_url} cannot reach')
except TimeoutError as te:
raise ImageDownloadTimeoutException(f'{_to_download_url} download timeout')
except CustomException as ce:
raise ce
except Exception as e:
raise ConsumerAlgorithmUncatchException(tb.format_exc())
def execute(
self,
_to_download_url,
_oss_helper,
_timeout=30,
_image_size_threshold=10,
):
to_return_result = OrderedDict()
try:
response = requests.get(_to_download_url, timeout=_timeout)
data_stream = BytesIO(response.content)
m_image_file_buffer = data_stream.read()
request_image = cv2.imdecode(
np.frombuffer(m_image_file_buffer, np.uint8), -1)
if request_image is None:
raise ImageFormatNotSupportException(
f'image:{_to_download_url} format not support,cannot decode by opencv'
)
if len(request_image.shape) == 3:
image_h, image_w, image_c = request_image.shape
else:
image_h, image_w = request_image.shape[:2]
image_c = 1
if _image_size_threshold is not None and request_image.nbytes < _image_size_threshold * 1024:
raise ImageFileSizeAbnormalException(
f'image:{_to_download_url} is small than threshold,it may not be a normal picture'
)
file_name = get_uuid_name()
oss_path = os.path.join(get_date_string(), file_name)
# 存储原始图像
saved_path = _oss_helper.upload_image_file('downloaded-image',
oss_path,
request_image,
_enable_compress=False)
to_return_result['bucket_name'] = 'downloaded-image'
to_return_result['saved_path'] = saved_path
to_return_result['image_height'] = image_h
to_return_result['image_width'] = image_w
to_return_result['image_channel'] = image_c
return to_return_result
except requests.exceptions.Timeout as te:
raise ImageDownloadTimeoutException(
f'{_to_download_url} download timeout')
except CustomException as ce:
raise ce
except Exception as e:
raise ConsumerAlgorithmUncatchException(tb.format_exc())
def __init__(self, _is_test):
self.logger = logging.getLogger(f'operator [{self.name}]')
if _is_test:
self.logger.setLevel(logging.DEBUG)
else:
self.logger.setLevel(logging.INFO)
# 会在工作目录下面创建文件夹logs，用于装载日志
os.makedirs('logs', exist_ok=True)
date_string = get_date_string()
log_format = f'%(asctime)s : %(levelname)s : %(process)d: %(thread)x: {self.name}: line %(lineno)d: %(message)s'
# 每天存储一个文件，存储31天，且文件名包含启动日期
log_file_handler = handlers.TimedRotatingFileHandler(
f'logs/{self.name}_{date_string}启动.log',
encoding='utf-8',
when='D',
interval=1,
backupCount=31)
log_file_handler.setFormatter(logging.Formatter(log_format))
self.logger.addHandler(log_file_handler)
self.is_test = _is_test
