Affordance Network Evaluation Script

[HUmutOzdemir]

Hi, I saw your work and wanted to examine it because it caught my attention. There was one thing that came to my mind during my review. As far as I understand, the affordance estimation network makes predictions in 2 separate channels for each affordance. One of the channels estimates the affordance probability and the other estimates the interaction probability. How do you make the final affordance estimate with these two estimates? I saw two different codes written for this.

Which one of these code snippets gives the final estimate? Or is there any other calculation strategy that you are using? And which threshold value did you use to convert estimation probability to estimated class value? Thank you in advance.

[Tushar-N]

The first one (def get_processed_affordances) here. The second function is just to accentuate the affordance maps for visualization, but not for evaluation.

[paulawulkop]

Thanks for the clarification. I have a quick follow up question: Which threshold value do you use to convert the float output from get_processed_affordances into a binary classification value?

[Tushar-N]

Threshold is just 0.5.