freesouls / caffe

Caffe For FaceNet: An Extension For Face Recognition

Caffe For FaceNet

Modified Caffe Framework For FaceNet

What is FaceNet?

Link: CVPR 2015 FaceNet: A Unified Embedding for Face Recognition and Clustering

Notes:

1. thought I implemented both offline and online triplet_loss_layer, but the loss does not decrease when training, If you can point out where is wrong, please let me know.([e-mail]) THANKS in advance.
2. the test codes I wrote in test_triplet_layer.cpp is not correct, so it will not pass when you run "make runtest" in shell, just ignore it. :)

Updates

#####15.7.26

• add TripleLossLayer class in loss_layers.hpp // empty class for now
• add triplet_layer.cpp and triplet_layer.cu //emtpy files, to be updated

#####15.8.13

• add message TripleLossParameter in src/caffe/proto/caffe.proto

#####15.8.14

• implemented Forward_cpu, Backward_cpu, SetUpLayer in src/caffe/layers/triplet_layer.cpp

#####15.8.15

• implemented Forward_gpu, Backward_gpu in src/caffe/layers/triplet_layer.cu
• and can be compiled successfully

#####15.8.16

• need further testing & fix bugs

Caffe

Caffe is a deep learning framework made with expression, speed, and modularity in mind. It is developed by the Berkeley Vision and Learning Center (BVLC) and community contributors.

Check out the project site for all the details like