A Deep Face Identification Network Enhanced by Facial Attributes Prediction

fusion layer :两个特征混淆，混淆的方法是使用 克罗内克积 第一如下：

损失函数：

Face Identification: 

we calculate the similarity between

each of the images in the gallery set and given image from
the probe set, and then rank these images based on the obtained
similarities. In face identification, the gallery set
should contain at least one image of the same identity. We
evaluate our model by using rank-1 identification accuracy
as well as Cumulative Match Characteristics (CMC) curves.
CMC is a rank-base metric indicating the probability of the
correct gallery image that can be found in the top k similar
images from the gallery set.

Facial Attribute Prediction: 

We leverage identity facial

attributes as an auxiliary modality for improving face identification
performance. Identity facial attributes are invariant
attributes which remain same from different images of
a person. For example, gender, nose and lips shapes remain
the same in different images of a person; however, attributes
such as glasses, mustaches, or beards may or may not exist
in different images of a person. We discard such attributes
in our model because we look for robust as well as invariant
facial attributes. Identity facial attributes in CelebA dataset
are listed as follows: narrow eyes, big nose, pointy nose,
chubby, double chin, high cheekbones, male, bald, big lips
and oval face . We evaluate our attribute predictor by using
accuracy metric.