How to decide value of start and end?

facebookresearch / odin

A simple and effective method for detecting out-of-distribution images in neural networks.

Other

530 stars 102 forks source link

How to decide value of start and end? #9

Open Arshita27 opened 5 years ago

Arshita27 commented 5 years ago

https://github.com/facebookresearch/odin/blob/64e97962ccaed1fe979f43a089c0feb4d8b002fd/code/calMetric.py#L62

How exactly is the value of 'start ' and 'end' chosen after performing temperature scaling?

Is it dependent on Temperature?
Is it dependent on the number of classes?

sam575 commented 5 years ago

In case if you have not figured it out yet. As per my understanding,

The start value is 1/num_classes.
After temperature scaling softmax scores are bound by a value less than 1. So, I assigned end as the rounded decimal of the max of softmax scores in confidence_Our_Out.txt and confidence_Our_In.txt.

cyyuu-bot commented 1 year ago

In case if you have not figured it out yet. As per my understanding,

The start value is 1/num_classes.

After temperature scaling softmax scores are bound by a value less than 1. So, I assigned end as the rounded decimal of the max of softmax scores in confidence_Our_Out.txt and confidence_Our_In.txt.

what about '100000' in 'gap=(end-start)/100000'?how to decide?