Face Detection and Comparison in DL4J

Hello! :wave:

We’re building a face detect model with DL4J, and have set up the FaceNet example from the Convolutional Networks tutorial.

We’re quite new to DL4j, and have only figured out how to use the model as a classifier (i.e. this photo contains a face).

We’d like to be able to crop a face within a larger photo (segmentation), and compare two faces to get a similarity score.
Is it possible to use the CNN (linked above) for segmentation, or will it require different architecture, or training on image segments?

Any advice will be appreciated. Thanks in advance!

The link you mentioned is about face recognition or using FaceNet to classify different faces. If you want to crop the face, you may use or train a yolo/ssd/r-cnn model first. The object detection model output the ROI which means the face area in the image. And then using FaceNet to classify the face into a category/person. But if there are too many faces to recognize, you may just use FaceNet/VGGFace to generate the feature vector, and then calculate the similarity score between them and the standard faces’ feature.To some extent, it’s like a image retrieval problem. Just FYI.