siqueira-hc/Efficient-Facial-Feature-Learning-with-Wide-Ensemble-based-Convolutional-Neural-Networks

Hi, can i train a binary problem?

Angel113110 opened this issue · 2 comments

Angel113110 commented

Dear author, thanks for sharing so wonderful work. I have a question to ask you. In my case, I only focus on smile expression. Can I train a binary problem based on your work? Can you give some help or advice? Thanks!

siqueira-hc commented

Dear Angel113110,

Thank you very much for your message and your interest.

There are several ways to use our model in this case, for example:

  1. Since smiling is strongly related to happy facial expressions in public datasets, you could directly use our model for facial expression recognition and post-processing happy classifications in order to detect smiles.

  2. The last layer of ESR-9 uploaded in this repository has two neurons (units) responsible for continuous affect perception. You could re-train this layer for your binary problem on your dataset using the cross-entropy loss function.

  3. The second last layer of ESR-9 uploaded in this repository has eight neurons (units) responsible for discrete emotion perception. You could use them as features to train another model for smile detection.

Best,
Henrique Siqueira

Angel113110 commented

Dear author. If I want to use the second way to train a binary task. How to do with it? I'm new in FER field. Can you say more details?