yuanmingqi / UniNet

Towards More Accurate Iris Recognition Using Deeply Learned Spatially Corresponding Features

UniNet

Towards More Accurate Iris Recognition Using Deeply Learned Spatially Corresponding Features

Project structure

This project is organized as follows:

• dataset: contains the training data, validation data and test data.
• reference: contains the reference materials.
• snapshots: for saving model weights.
• static: contains the pretrained model (MaskNet) and so on.
• data.py: for constructing the data generator.
• eval.py: for calculating TAR, FAR, EER.
• loss.py: the implementation of the Extended Triplet Loss.
• mask.py: for predicting masks for all the images in dataset using the pretrained MaskNet.
• match.py: functions for the iris matching (e.g. Hanmming distance).
• model.py: model class of the UniNet.
• network.py: the implementation of the FeatNet and the MaskNet.
• train.py: training file.

Packages required

The moudle below is required for the project:

• PyTorch
• Pandas
• NumPy
• PIL

Dataset preparation

For conducting training, your dataset should be organized as follows:

Train.txt

.dataset/Train/person1_001 0 
.dataset/Train/person1_002 0
...
.dataset/Train/personN_001 N
...