2

I am trying to extract the pixel wise probabilities for each semantic class present in an Image when performing semantic segmentation using Google's DeepLab V3+. I am following the demo given here - https://github.com/tensorflow/models/tree/master/research/deeplab .

I had the following questions - 1. How to get class wise probability scores for each Semantic class in Image Segmentation using Google's DEEPLAB V3+. 2. Also what is the threshold for each class?

Would be grateful for any help in this direction.

Thank you :)

Given link below has the graph definition for the network DeepLab V3+ I'm using - https://paste.fedoraproject.org/paste/pbDVRLOaUgtZmT-rbEVmbQ

Sid Gairola
  • 101
  • 1
  • 6

1 Answers1

3

You can get the probability by using the following: Deeplab code

predictions[output + PROB_SUFFIX] = tf.nn.softmax(logits)

EDIT: The latest deeplab now provides the probabilites as well.

Neeraj Gulia
  • 640
  • 8
  • 24