FIW supports laboratory style evaluation protocols, Kinship Verification (i.e., one-to-one) and Family Classification (i.e., one-to-many).
Data and resources (e.g., source code, trained models, baseline features, etc.) will be made available soon!
Data and resources (e.g., source code, trained models, baseline features, etc.) will be made available soon!
Kinship Verification
Kinship verification tests were run on previously tested pairs such as siblings, mother – daughter, father – son, etc. along with four newly introduced multi generational pairs between grandparents and grandchildren in the FIW database. With the Fine-Tuned CNN model, we saw an average of a 4% increase in accuracy from the VGG-Face model. The benchmark tests were completed by randomly selecting an equal number of positive and negative pairs for each family, then computing cosine similarity for each pair.
Family Classification
By fine-tuning the VGG-Face model, we achieved an average of a 1% increase in accuracy. This was accomplished by following the 5-fold cross-validation protocol with no family overlap on families with more than five members. Folds were then made up of one member for each family and multi-class SVM was used to model VGG-Face features for each family.
|