Siamese neural network for the website ucoin.net
A model based on two convolutional neural networks (CNNs) with equal weights and architecture.
The model accepts two images as input and outputs the images similarity coefficient ranging between 0 and 1.
Differs from similar neural networks in using a small dataset.
- Python 3.7
- Jupyter
- Tensorflow
- Keras
- Keras-tuner
- Numpy
- Pillow
images_prep
- images preparationdata_prep
- data preparationmodel_tuner
- hyper parameters tuningsiamese_nn
- network creation and learningimages
- directory for user’s imagesdata
- directory with prepared data
To train the model on your images, create a cascade of directories, each of which contains a set of images of the same class/type.
Example:
images
|_____s1
| |_____1.jpg
| |_____2.jpg
| |_____ ...
| |_____10.jpg
|_____s2
|_____s3
|_____ ...
|_____s40