output name getting via getUnconnectedOutLayersNames is different from description of OpenVINO model #20380
Description
System information (version)
- OpenCV => OpenCV+OpenVINO Windows package 4.5.2 or 4.5.3
- Operating System / Platform => Windows 64 Bit
- Compiler => Visual Studio 2019
- Python => 3.7.9
Detailed description
output name getting via getUnconnectedOutLayersNames is different from description of OpenVINO model
person-vehicle-bike-detection-2004
currently found three models, the doc page listed above said it has two outputs boxes and labels
but there's not just two when we try getting output via dnn.getUnconnectedOutLayersNames
then if we pass the output to forward of DNN, error occurred
model person-detection-0203
====OutputNames====
['1095_1096.0', '1312_1313.0', '1529_1530.0', '1746_1747.0', '1963_1964.0', '223
4_2235.0', 'boxes', 'labels']
====OutputNames====
Traceback (most recent call last):
File "dnn.py", line 58, in <module>
outs = net.forward(outNames)
cv2.error: [ PARAMETER_MISMATCH ] Failed to set output Blob. Dimensions mismatch
.
if we specify output ['boxes', 'labels'] , same error
model person-detection-0203
====OutputNames====
['1095_1096.0', '1312_1313.0', '1529_1530.0', '1746_1747.0', '1963_1964.0', '223
4_2235.0', 'boxes', 'labels']
====OutputNames====
Traceback (most recent call last):
File "dnn.py", line 58, in <module>
outs = net.forward(outNames)
cv2.error: [ PARAMETER_MISMATCH ] Failed to set output Blob. Dimensions mismatch
.
if we specify output ['boxes'] , there's no error, but getting different result when running multi times
the input image is the same one
Steps to reproduce
import cv2
# OpenCV+OpenVINO Windows package
# https://github.com/opencv/opencv/wiki/Intel's-Deep-Learning-Inference-Engine-backend#opencvopenvino-windows-package-community-version
model_root = '/intel'
# https://docs.openvinotoolkit.org/latest/omz_models_model_person_vehicle_bike_detection_2003.html
prefix = 'person-vehicle-bike-detection-2004'
size = (448,256)
# https://docs.openvinotoolkit.org/latest/omz_models_model_face_detection_0206.html
prefix = 'face-detection-0206'
size = (640,640)
# https://docs.openvinotoolkit.org/latest/omz_models_model_person_detection_0203.html
prefix = 'person-detection-0203'
size = (864,480)
print('model {}'.format(prefix))
net = cv2.dnn.Net_readFromModelOptimizer(
model_root + '/{}/FP32/{}.xml'.format(prefix,prefix),
model_root+ '/{}/FP32/{}.bin'.format(prefix,prefix))
net.setPreferableBackend(cv2.dnn.DNN_BACKEND_INFERENCE_ENGINE)
net.setPreferableTarget(cv2.dnn.DNN_TARGET_CPU)
frame = cv2.imread('./test.jpg')
blob = cv2.dnn.blobFromImage(frame, 1.0, size, swapRB=False, crop=False)
outNames = net.getUnconnectedOutLayersNames()
print('====OutputNames====')
print(outNames)
print('====OutputNames====')
# outNames = []
# outNames.append("boxes")
# outNames.append("labels")
net.setInput(blob)
outs = net.forward(outNames)
print('====OriginalResult====')
print(outs[0][0])
print('====OriginalResult====')