Reading annotated images

[kaulanishasj]

Hi,
I just started exploring your codebase. I am trying to run segnet for my own project in which the images have fewer number of classes. When I try to just simply read one of the annotated images: '0001TP_006690.png' in the trainannot directory, I get this:

[[ 0.00392157 0.00392157 0.00392157 ..., 0.00392157 0.00392157
0.00392157]
[ 0.00392157 0.00392157 0.00392157 ..., 0.00392157 0.00392157
0.00392157]
[ 0.00392157 0.00392157 0.00392157 ..., 0.00392157 0.00392157
0.00392157]
...,
[ 0.01568628 0.01568628 0.01568628 ..., 0.04313726 0.04313726
0.04313726]
[ 0.01568628 0.01568628 0.01568628 ..., 0.04313726 0.04313726
0.04313726]
[ 0.01568628 0.01568628 0.01568628 ..., 0.04313726 0.04313726
0.04313726]]

The way I tried to read the png is by:
import matplotlib.image as img
image = img.imread('0001TP_006690.png')
print(image)

If i am not wrong, arent the classes supposed to be a whole number like 1,2..11(since there are 11 classes) instead of decimnal point numbers? or is there a conversion happening that I am not aware of.

[tkuanlun350]

I think matplotlib rescaled the data by itself. According to the document, "Matplotlib has rescaled the 8 bit data from each channel to floating point data between 0.0 and 1.0".(https://matplotlib.org/users/image_tutorial.html)

Maybe you can try to use my input reader in Inputs.py, it will use multi-thread and deocde png file by tensorflow library.