I am working on a project of information extraction from images with OpenVino and Python3. Because of the extension of the project, it was decided that our face detection module would work on an independent process with the support of a multiprocessing module.
Unfortunately, I am not managing to make the face detection work within this framework, because the infer() method from IeCore seems to block (there are no primary implementation problems because I can make the mentioned method works properly when using treading module instead of the process).
So, any help on how to work around this issue is welcome.
Bellow, the main elements of the code that I am trying to run with the multiprocessing module.
import os
import cv2
import time
import numpy as np
from pathlib import Path
from multiprocessing import Process
from modules.face.abstract.abcopenvino import ABCOpenVino
from modules.utils.blob2frame import blob2frame
from openvino.inference_engine import IECore, Blob, TensorDesc
YAXIS = 0
XAXIS = 1
class FaceDetection(ABCOpenVino):
def __init__(self, frame_queue, face_queue):
super().__init__()
if not hasattr(self, 'confidence'):
raise ValueError('Class attribute confidence must be defined')
self._smoothed = np.zeros((4, )).astype('float32')
self.frame_queue = frame_queue
self.face_queue = face_queue
ie_core_handle = IECore()
self._network = ie_core_handle.read_network(self.file_xml, self.file_bin)
key = next(iter(self._network.input_info))
self.tensor_description = self._network.input_info[key].tensor_desc
self._executable_network = ie_core_handle.load_network(self._network, device_name='CPU', num_requests=1)
self.inference_request = self._executable_network.requests[0]
self.input_blob_name = next(iter(self.inference_request.input_blobs))
self.output_blob_name = next(iter(self.inference_request.output_blobs))
self.p = Process(target=self._process_exec, args=())
self.p.name = '{}_{}'.format(__name__, 0)
self.p.daemon = True
self.p.start()
time.sleep(0.01)
# With treading it works
#import threading
#t = threading.Thread(target=self._process_exec, args=())
#t.name = '{}_{}'.format(__name__, 0)
#t.daemon = True
#t.start()
#time.sleep(0.01)
def _process_exec(self):
self.myprint("Running! pid: {}".format(os.getpid()))
while True:
if self._stop:
self.myprint('Stopping! pid: {}'.format(os.getpid()))
time.sleep(0.05)
break
elem = self.frame_queue.get()
if elem:
frame, = elem
face_bbox_lst = self._network_infer(frame)
if face_bbox_lst:
elem = (frame, face_bbox_lst, )
self.face_queue.put(elem)
self.myprint('Done')
return
def _network_infer(self, frame):
N, C, H, W = self.tensor_description.dims
hwc = cv2.resize(frame, (H, W,)).astype('float32')
chw = hwc.transpose((2, 0, 1))
input_blob = Blob(self.tensor_description, chw)
self.inference_request.set_blob(blob_name=self.input_blob_name, blob=input_blob)
self.inference_request.infer()
output = self.inference_request.output_blobs[self.output_blob_name].buffer
return output
def terminate(self):
self._stop = True
