Image process should be sub 127.5 then div 127.5

[xiaoxiong1006]

Sorry for my English. Refer in https://tfhub.dev/tensorflow/lite-model/deeplabv3/1/metadata/2,the origin image process should be sub 127.5 then div 127.5.The code in line 2 of setcion 6 should like this:

np_res_im = ((np_res_im-127.5)/127.5)).astype('float32')

displace

np_res_im = (np_res_im/255).astype('float32')

After do it,the accuracy will improve.

[joonb14]

@xiaoxiong1006 You are the best!
Since I don't have jupyter environment anymore, I will replace it latter!
Thank you!