topic facial-landmarks-35-adas-0002 output shape value in Embedded Intel® Core™ Processors

facial-landmarks-35-adas-0002 output shape value

KKar01 — Tue, 21 Jan 2020 00:13:09 GMT

Hello, in the documentation, the output shape of the landmark is described as The net outputs a blob with the shape: [1, 70], containing row-vector of 70 floating point values for 35 landmarks' normed coordinates in the form (x0, y0, x1, y1, ..., x34, y34). Howevr in my Python code I get shape as (1,57330). Can you tell me what I am doing wrong?

Lmk = cv2.dnn.readNet('facial-landmarks-35-adas-0002.xml',

'facial-landmarks-35-adas-0002.bin')

# Specify target device.

Lmk.setPreferableTarget(cv2.dnn.DNN_TARGET_MYRIAD)

Lmk.setInput(cv2.dnn.blobFromImage(frame, size=(672, 384), ddepth=cv2.CV_8U))

lmkOut = Lmk.forward()

print (lmkOut.shape)

Re: facial-landmarks-35-adas-0002 output shape value

KKar01 — Tue, 21 Jan 2020 04:19:56 GMT

I finally figure this out and have two important observation - can you help me answer by

1) the set input size should be (60,60) to return a shape of 70 - so the line should be Lmk.setInput(cv2.dnn.blobFromImage(frame, size=(60,60), ddepth=cv2.CV_8U)) . question is why 60 works?

2) i have a wide angle camera in my raspberry Pi and with this the face comes small relative to the whole image --> in this case the key point prediction is still within the bounding box , but do not align to the eye nose position. Do you know why this is happening?

Can you please help answer the above questions?

Re: facial-landmarks-35-adas-0002 output shape value

CarlosAM_INTEL — Tue, 21 Jan 2020 04:32:26 GMT

Hello, @KKar01:

Thank you for contacting Intel Embedded Community.

The consultations stated in this thread should be addressed through the channels listed at the following website:

https://software.intel.com/en-us/forums/intel-distribution-of-openvino-toolkit

Best regards,

@Mæcenas_INTEL.