How to create a color map image of uncertainty with CNN's deep output?

Question

I design CNN deep classifiers to detect urban features. Most of the time, my network classifies and segments the construction properly, but many times it is confused due to lighting / similar appearance, etc. With other objects.

I want to create a color map along with a segmented image that can represent as a specific classifier? I used softmax with loss for training network.

layer {
  name: "score"
  type: "Deconvolution"
  bottom: "pool_3"
  top: "score"
  convolution_param {
    num_output: 2
    bias_term: false
    pad:2
    kernel_size: 8
    stride: 4
  }
}

I expect a result similar to this color map image:

My problems

• How to calculate the uncertainty?
• How to handle negative values ​​when calculating uncertainty?

Note. Currently, I can get a color map using entropy.

Answer 1

, , .

AI StackExchange:

. , ConvNet . , , - ?

, - , (, ) . , , , .