Using VGG16 to Recognize Faces. Research with Thay Viet (thviet79), Tra Oliver (Traoliver)
Clone mtcnn project from https://github.com/ipazc/mtcnn/tree/master/mtcnn to detect human faces.
On this project, we trained model to recognize Face on 3 dataset sample AT&T, Yale and one outer dataset ( students - 26 person - 11-20 images/person)
Preview data
Yale:
AT&T:
Student:
- This dataset was crawled from facebook (for research only) and cropped to satisfy conditions 1 image had only 1 person . After that we use MTCNN to extract faces from there dataset. The difficult of there dataset was it contains multi part and angle of face.
Here is the result after training:
AT&T(40 person): We trained AT&T dataset by divide to 2 part 40% for training and 60% for validation. Train time 7 minute and maximum validation accuracy it 95% after 20 times trained.
Yale (15 person): We trained AT&T dataset by divide to 2 part 40% for training and 60% for validation. Train time 5 minute and maximum validation accuracy it 100% after 20 times trained.
AT&t + Yale dataset (55 person):
We trained AT&T dataset by divide to 2 part 40% for training and 60% for validation. Train time 14 minute and maximum test accuracy it 95.76% after 20 times trained.
Student:
Train time 14 minute 3 hour
Id - 17103100397
Id - 17103100409
We also should tested there model with a person non-trained and it respond well
