Face Recognition

Computer Vision and Intelligence Group, IIT Madras

We implement an experimental setup with face detection and recognition. This has been used for our purposes with the following aims:

• Swapping multiple detectors and feature extractors for facenet.
• Multi GPU and distributed support
• Frozen graph support with quantisation.

Primarily, we use this in two use cases:

• High accuracy: SSD or FRCNN detectors with Inception-Resnet feature extractors.
• CPU optimised FPS: SSDlite mobilenet with mobilenet V2 extractors (this is covered in getting started).

We have experimented with multiple classifiers for error free student identification.

Sample Results

Contents

• Getting Started
• Model Zoo
• Installing Dependencies

Facenet Docs

• David Sandberg's Implementation
• Retraining Facenet
• Recognition with Facenet
• Facenet Wiki

Object Detection Experimental Setup

• Object Detection Pipeline

To Do

• [ ] TF-Estimator based scalable train file.
• [x] SSDLite based detector
• [x] Mobilenet models for facenet
• [x] Angular, Focal and triplet losses.
• [ ] Inference on Singular Videos.
• [ ] DALI, Tensor RT for faster inference.
• [ ] S3D support for detection.
• [ ] Experiments with weight tying.
• [ ] Results Section
• [ ] Take a look at https://github.com/alexattia/ExtendedTinyFaces for large scale face detection.

Dependencies

• Python 3.4+
• Tensorflow 1.7+
• Opencv 3.3.1+

Pipeline

Image -> FaceDetection -> CroppedFace -> FaceRecognition -> Descriptor(128D) -> FaceClassifier -> Name

Credits

FaceRecognition(FaceNet)

TensorFlow implementation of the face recognizer described in the paper "FaceNet: A Unified Embedding for Face Recognition and Clustering". Ref. https://github.com/davidsandberg/facenet