This sample demonstrates real-time face recognition on Android. The project is based on the FaceNet now. FaceNet output the image's face feature embedding vector. We store only one picture of the same person. We compare the detected real-time face with the FaceDB's face feature by computing two faces' Euclidean distance. If The distance is smaller than THRESHOLD, and it is the minimum compared result, we consider they are the same person.
The project is heavily inspired by
- You can click the '+ 'button ,then add the new person's face image and name that you want to recognize.(Source)
- You can click one person's name in that list to delete his face data.
from davidSandberg' s facenet
| Model name | LFW accuracy | Training dataset | Architecture |
|---|---|---|---|
| 20180402-114759 | 0.9965 | VGGFace2 | Inception ResNet v1 |
- Use more accurate, lighter and faster face recognition model
- Use better and faster face detector
- ...