Error running instance-segmentation-security inference via OpenCV's dnn (with IE)

Hi,

I’m running into an error inferring instance-segmentation-security-1039 using OpenVino via OpenCV’s dnn module: `cv2.error: [ PARAMETER_MISMATCH ] Failed to set output Blob. Dimensions mismatch.

This my basic test:

import cv2 as cv
import numpy as np
model_name = 'instance-segmentation-security-1039'
net = cv.dnn.readNetFromModelOptimizer(model_name+'.xml', model_name+'.bin')
print(net.dump())
cap = cv.VideoCapture("test_video.mp4")
# instance-seg = 480
(n, c, h, w) = (1, 3, 480, 480)
# yolact = 550
# (n, c, h, w) = (1, 3, 550, 550)

while cv.waitKey(1) < 0:
    hasFrame, frame = cap.read()
    
    if not hasFrame:
        break
    
    blob = cv.dnn.blobFromImage(cv.resize(frame, (w, h)))
    net.setInput(blob)
    out = net.forward()

    cv.imshow('OpenVINO', frame)

This is using OpenCV 4.5.3-openvino (which shipped with openvino_2021.4.689).

I can run the instance_segmentation_demo using the IE directly just fine.

Is it possible to run the same demo via OpenCV ? If so, how ?

Thank you so much for your time, George

[alalek]

4.5.4-dev should work with OpenVINO 2021.4.689 package (2021.4.1 release)

Inference Engine: YES (2021020000 / 2021.4.1) 2021020000 / 2021.4.1

We have here unaligned declared and used IE versions (OpenCV uses OpenVINO product version). Current value says that we using OpenVINO 2021.2.0 which is not correct. We need to replace: 2021020000 => 2021040100 (INF_ENGINE_RELEASE CMake variable) to properly build code (e.g., compile this code at least).

Perhaps this value is stored in CMakeCache.txt somewhere from previous old cmake runs. There is general CMake rule that we need to drop its caches if we change/upgrade dependencies. So, to avoid other issues it is better to build from the clean build directory.

[alalek]

PARAMETER_MISMATCH

relates https://github.com/opencv/opencv/pull/20410

This patch will be released with OpenVINO 2022.1 (2021.4.x LTS updates are out of scope for now)

Also you can build OpenCV from source code with enabled InferenceEngine: https://github.com/opencv/opencv/wiki/Intel's-Deep-Learning-Inference-Engine-backend#build-opencv-from-source