emotion result varies between 2 iteration for the same image

[chuyee]

Hi,

I did a small modification to make poc.py read from an image file instead of continuous video capture. However I got different result each time I predict the same image. i.e.

frame = cv2.imread('test.jpg')

# Predict result with network
result = network.predict(format_image(frame))
print([ '%.2f' % i for i in result[0] ])

The results are as below:

['0.13', '0.17', '0.13', '0.17', '0.13', '0.17', '0.11']
['0.09', '0.22', '0.16', '0.15', '0.13', '0.13', '0.12']
...

Is this expected or I did something wrong?

Thanks, -yi

[maxicastro]

Hi @chuyee, i done the same change to the code and have the same results: got different result each time i run. On the other hand, have you been able to achieve better results? any are more than 20%, the same here.

[chuyee]

nope, I gave it up and switched to something else...