deven96 / face-similarity

Face-similarity CNN using Tensorflow Eager execution.

Face-similarity

Face-similarity CNN using Tensorflow Eager execution.

Medium article: How to train your own FaceID ConvNet using TensorFlow Eager execution

This implementation uses DenseNets with contrastive loss.

Reference Papers:

• Densely Connected Convolutional Networks
• Dimensionality Reduction by Learning an Invariant Mapping

Dependencies:

• Python 3.6.x
• Tensorflow 1.10.1

Inference

Google Colaboratory

• Just open the inference.ipynb and select the option to open on Colab.

Running locally

1- Download the pre-trained model using the following link.

• Place the tboard_logs folder in the root folder of the project.

• Tensorflow pre-trained model

2- Download the following test dataset (TfRecords format).

• Place the dataset folder in the root folder of the project.

• Download test dataset

3- Run the jupyter notebook inference.ipynb

• Run the notebook. Adjust the dataset paths accordingly.

Results