[repo in process]
Working on a FaceID-like (low fidelity) app that unlock your computer. This project uses deep learning through three main steps :
- Warming up with MNIST
- Detecting faces
- Verifying face
- Unlocking computer via shell script.
The first part has been done on MNIST dataset to understand the concept of Siamese Networks: same networks are used to turn images into embeddings (128 dimension encoding vectors). Each embedding vector represents (encodes) an image. The more 2 images are different, the larger the difference between their embedding vectors. Training has been performed with Online Triplet Mining.
This first project on MNIST gives visualizations in 2D space thanks to the TSNE algorithm. There are two results:
- the first shows how the model takes apart each handwritting digits from each other,
- the second shows how the model takes apart all the digit (label is 0) regarding a chosen one (here the 7).