facial-landmarks-35-adas-0001 input description

[banderlog]

It stated:

The blob is constructed from a 60x60 pixel BGR image with subtracted mean of (120, 110, 104) and scale factor of 0.0039.

And I have spent a lot of time, to figure out, that right parameters must come without mean and scale, or you'll get warped/distorted result.

# original, not working
blob = cv2.dnn.blobFromImage(face, 0.0039, (60, 60), mean=[120, 110, 104], swapRB=False)

# working
blob = cv2.dnn.blobFromImage(face, 1, (60, 60))
# also working, but worser
blob = cv2.dnn.blobFromImage(face, 1, (60, 60), mean=[120,110,104], swapRB=False)

[snosov1]

Hey, @banderlog !

Sorry, I'm having hard time to extract an actionable item from your description.

• Do you imply that the documentation is misleading (e.g. simply eliminating the mean/scale text from the document will avoid the confusion)?
• Is there an actual code in this repo that needs to be changed?

[banderlog]

My bad, @snosov1
Yes, I'm implying that the documentation is misleading and simple 1 string change will help to avoid confusion.

[snosov1]

I'm with you on that one, care to submit a PR?

[snosov1]

Sorry, I see you've already created one #48 =)

[snosov1]

Not relevant anymore with 0002 replacing 0001 model