Siamese Neural Networks

tensorflow implementation of (Koch et al. 2015)

Instructions

• Set Up Virtual Environment (Optional but Recommended)

python -m venv venv
source venv/bin/activate 

• Clone the repository

git clone https://github.com/thlurte/SiameseNet-tensorflow.git

• Change into the directory of the repository

cd SiameseNet-tensorflow

• Install Dependancies

pip install -r requirements.txt

• To Train the model

python main.py --data_dir <path-to-dataset>

Architecture

The model consists of a sequence of convolutional layers, each of which uses a single channel with filters of varying size and a fixed stride of 1.

Activation

The network applies a ReLU activation function to the output feature maps, optionally followed by maxpooling with a fliter size and stride of 2.

Loss Function

Binary cross entropy is used as loss function to calculate the error.

Optimizer

Adam is used to find the global minima in this architecture.

Weight Initialization

All network weights are initialized in the convolutional layers from a normal distribution with zero-mean and a standard deviation of $10^{−2}$. Biases were also initialized from a normal distribution, but with mean 0.5 and standard deviation $10^{−2}$.

References

• Siamese-Neural-Networks-for-One-shot-Image-Recognition
• Siamese neural networks for one-shot image recognition