- 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>
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.
The network applies a ReLU activation function to the output feature maps, optionally followed by maxpooling with a fliter size and stride of 2.
Binary cross entropy is used as loss function to calculate the error.
Adam is used to find the global minima in this architecture.
All network weights are initialized in the convolutional layers from a normal distribution with zero-mean and a standard deviation of