Face Net In Xiao Yu

Compatibility

The code is tested using Tensorflow 1.2 under Ubuntu 16.10 with Python 3.5

Journey

Date	Update
2017-12-1	Completed train Triplet Loss code. It can be used to train different CNNs Model
2017-12-6	Completed the pipeline system
2017-12-15	Test the Pre-trained model (Inception-ResNet-v1 Model) in Xiao Yu Pipeline data set

Codes

• pipeline Fold : Pipeline In Xiao Yu
• model Fold: The CNNs models used to extract the image features

Training data

• Small data set: LFW
• IMDB data set (Cropped): IMDB_crop
• Wiki data set (Cropped): WIKI_crop

Also we can use the shell script to download the trainin data set:

./download.sh

Pre-trained models

Model name	LFW accuracy	Training dataset	Architecture
20170511-185253	0.987	CASIA-WebFace	Inception ResNet v1
20170512-110547	0.992	MS-Celeb-1M	Inception ResNet v1

Todos

• Begin the project and coding the major code heavily inspired by FaceNet
• Try to train the MobileNet with triplet loss
• Test the pre-trained model (Face net) in Pipeline
• Use Pre-trained model as data extraction and train gender with last full connection after data extraction, using Softmax function as training function