Deep Metric Learning
Let’s say we want to create a model that can do face recognition(Face Identification and face verification). We CAN use traditional deep learning since it can perform really well but it requires a lot of training data. This means we will need A LOT of images for each person.
Deep metric learning can be helpful with these kinds of tasks(such as one-shot learning). Predicting with very few training data. Metric learning is an approach focusing on a distance metric to distinguish similarity or dissimilarity between images.
Triplet Network
Triplet Network is one of the fundamental ideas where metric learning is performed.
By giving a set of 3 images,
- Anchor Image: The Base Image
- Positive Image: Image of the same person
- Negative Image: Image of a different person
The Network tried to minimize the distance between the embedded anchor image and the embedded positive image while maximizing the distance between the embedded anchor image and the embedded negative image.
