Shivsharma779/facenet-implementation

Implementation of paper: FaceNet: A Unified Embedding for Face Recognition and Clustering

FaceNet implementation

Problem satement

In this Project we implemented Google's FaceNet , that directly learns a mapping from face images to a compact Euclidean space where distances directly correspond to a measure of face similarity.

Presentation

Datasets

• Training Dataset
• Testing Dataset

Steps

• Install all datasets and models using

bash scripts.sh

• Start the training script

python3 ./Code/train.py

• Evaluate Results

python3 ./Code/eval_model.py

Model will be generated under the name checkpoint.pth

Results

• Summary

	Margin = 1.7	Margin = 1.9
kNN	75.5	79.55 (VggFace2)
Contrastive	88.23	89.33 (LFW)
Rank5	57.9(LFW) / 70.17 (VggFace2)	57.85 (LFW) / 72.33 (VggFace2)
Loss	0.1013 (LFW) / 0.0391 (VggFace2)	0.1867 (LFW) / 0.1111 (VggFace2)

• TSNE Plot of different identities in VGG Face 2 dataset

• RESNET50 RESULTS(MARGIN:0.7)

  • Accuracy:

  

  • Loss

  

• RESNET50 RESULTS(MARGIN: 1.9)

  • Accuracy:

  

  • Loss