For a fair comparison with other studies, authors are encouraged to use the following experimental protocols used in:
A two level 10-fold cross-validation scheme is used. Each time a test fold is separated, a 9-fold cross-validation is used to train/validate the system, and parameters are optimized without using the test partition. In the 9-fold cross-validation, one fold is separated for validation where the rest is used for training. There is no subject overlap between folds.
- Dibeklioglu, H., A.A. Salah, and T. Gevers, "Are You Really Smiling at Me? Spontaneous versus Posed Enjoyment Smiles," Proc. European Conference on Computer Vision (ECCV), pp.525-538, 2012.
- Dibeklioglu, H., T. Gevers, A.A. Salah, and R. Valenti, "A Smile Can Reveal Your Age: Enabling Facial Dynamics in Age Estimation," Proc. ACM International Conference on Multimedia, pp.209-218, 2012.
A two level 10-fold cross-validation scheme is used. Each time a test fold is separated, a 9-fold cross-validation is used to train/validate the system, and parameters are optimized without using the test partition. In the 9-fold cross-validation, one fold is separated for validation where the rest is used for training. There is no subject overlap between folds.
Seven different sets of folds (for different experiments) can be found in the package. "fold_all" is the superset of the other sets. The architecture of the folds can be shown as follows:
For any questions about the experimental protocols, you can send an email to h.dibeklioglu [at] cs.bilkent.edu.tr