Dnn module gives error in net.forward() (eltwise_layer.cpp: input[0][j] == input[i][j] in function getMemoryShapes)

  • OpenCV => (opencv-python)
  • Operating System / Platform => Ubuntu 18.04
  • Compiler => Python3.7

I have converted BiSeNet network to frozen .pb file. Then I tried to read and inference it with OpenCV. But got an error (inputs[0][j] == inputs[i][j] in function ‘getMemoryShapes’.)

I think the problem can be in GlobalAveragePooling2D layer, because if I run net.getUnconnectedLayerNames(), there are 4 GlobalAveragePooling2D layers.

1st thing to do: update to recent 4.5.1 (and try again)

Thanks for answering. I have done update and nothing changed.