Help for implementing SegNet for predicting Single class
ArunJ1 opened this issue · comments
Dear All,
I am trying to implement SegNet to detect Speedhumps in a given image. I have created my custom dataset, by capturing images of Speedhump and labeling them.
I have a few issues and could you please kindly guide me?
As suggested in the above link, I have set num_output in softmax loss layer as "2". I removed the ignore label also.
-
I have labeled my image dataset as follows
Pixel value 0 for Background
Pixel value 1 for Speedhump.
So, is it okay or I need to set Pixel value as 1 for Background and Pixel value 0 for Speedhump? If I need to do so, please guide me with any easy ways to change it quickly. -
Is it Mandatory to calculate and apply the class weighting?
I really appreciate any help regarding the above issues
Thanks in advance,
Arun