Training a Deeplearning model with 16 bits images and make inference in 8 bit images, is it possible?

user_9583035 0
Viewed 39

I am training a deeplearning model using for train images with 16-bits

My test dataset have images off same class but 8 bits.

For pre- processing i am doing :

For 16 bits:

(1./(255.*255.)* batch_x_img

For 8 bits:

(1./(255.)* batch_x_img

Unfortunately, I can't get a good result on my test set.

In case I split my 16-bits set into training and testing. I get good results.

If I split my 8-bit set into training and testing . I get good results.

But i want use my 16bits images just for train and my 8 bits images just for test.

What Should i do in the pre-processing to make it possible?

tensorflow python
1 Answers

I realized that i need For 16 bits:

int((1./(255.)* batch_x_img)
(1./(255.)* batch_x_img

For 8 bits:

(1./(255.)* batch_x_img
user_95830350
Related