classify model not work as well as python test?

[HolyShitKillingMachine]

trained my model with inception v1,and trained for about 30000 times.
got training accurancy 1,and val accurancy 0.91
test it with python script,work well.
when imported it in the project,and change the factor,mean_std,classifyimagesize,and changed with vuforia camera.
but it works not well as python works.
the problem is the camera image or the training pics?


and the result of the classify result is always changing very big,maybe it thinks this image is like the potato and pepper.

hope you can give some advices to improve the effects.

I prepare to capture images of potato,pepper,tomato with my iphone,and test it again.i do not know if it works well.

[Syn-McJ]

@HolyShitKillingMachine,

Your model produces weird results, recognition probabilities shouldn't be negative. But if it works for your case then ok.

Regarding performance: as I mentioned in the Readme, i'm not sure how to improve it.

[HolyShitKillingMachine]

The negative result maybe mean that the image does not like the classify,cause it is negative.
I am doing this according to the page of someone's blog!
Thanks for your advice!i will try to improve it!