search

batmanlab / Explanation_by_Progressive_Exaggeration

Explanation by Progressive Exaggeration
https://openreview.net/pdf?id=H1xFWgrFPS
MIT License
20 stars 3 forks source link

Figure 6 #4

Open prlz77 opened 4 years ago

prlz77 commented 4 years ago

Hi @sumedhasingla ,

I am trying to reproduce Figure 6 of the paper but I cannot find the images you use. Would you mind to tell me how to find them?

Thanks!!

Pau

sumedhasingla commented 4 years ago

Dear Pau, For Figure 6, one need to do following steps:

  1. Train a binary classifier to detect if a person is smiling or not. In doing so use a dataset that has the same number of men and women in both positive and negative classes.
  2. Train a new binary classifier to detect if a person is smiling or not. This time, all smiling faces are of men and all not smiling faces are of females.
  3. For both the classifiers, process the dataset through the classifier and save their outputs.
  4. Create bins or partitions based on the classifier output. Refer this
  5. Train two explainer models one for each classifier.
  6. Compare the output of the explainer on the same input image for the two explainers.

I hope I answered your question. Let me know if you need any more information.