raphael/siamese
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

updates readme

Browse Source
This commit is contained in:
raphael 2021-08-17 10:00:13 +02:00
parent 6298c177bb
commit 3ef8c74f2e
1 changed files with 6 additions and 63 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

69
README.md
View File

@ -1,30 +1,14 @@
# Siamese Neural Network for Keras # Siamese Neural Network for Keras
This project provides a lightweight, easy to use and flexible siamese neural network module for use with the Keras This project provides a lightweight siamese neural network module for use with the Keras
framework. framework.
Siamese neural networks are used to generate embeddings that describe inter and extra class relationships. The neural network compares two images and returns their similarity in a 0-1 float value.
This makes Siamese Networks like many other similarity learning algorithms suitable as a pre-training step for many The datasets used are the fruit-360 dataset, COCO 2014 and 2017 as well as some ImageNet data.
classification problems.
An example of the siamese network module being used to produce a noteworthy 99.85% validation performance on the MNIST # Installation
dataset with no data augmentation and minimal modification from the Keras example is provided.
## Installation - tensorflow
Create and activate a virtual environment for the project.
```sh
$ virtualenv env
$ source env/bin/activate
```
To install the module directly from GitHub:
```
$ pip install git+https://github.com/aspamers/siamese
```
The module will install keras and numpy but no back-end (like tensorflow). This is deliberate since it leaves the module
decoupled from any back-end and gives you a chance to install whatever backend you prefer.
To install tensorflow: To install tensorflow:
``` ```
@ -36,19 +20,7 @@ To install tensorflow with gpu support:
$ pip install tensorflow-gpu $ pip install tensorflow-gpu
``` ```
## To run examples - If you use tensorflow-gpu, you'll have to install Cuda and CuDNN.
With the activated virtual environment with the installed python package run the following commands.
To run the mnist baseline example:
```
$ python mnist_example.py
```
To run the mnist siamese pretrained example:
```
$ python mnist_siamese_example.py
```
## Usage ## Usage
For detailed usage examples please refer to the examples and unit test modules. If the instructions are not sufficient For detailed usage examples please refer to the examples and unit test modules. If the instructions are not sufficient
@ -114,35 +86,6 @@ siamese_network.fit(x_train, y_train,
epochs=epochs) epochs=epochs)
``` ```
## Development Environment
Create and activate a test virtual environment for the project.
```sh
$ virtualenv env
$ source env/bin/activate
```
Install requirements
```sh
$ pip install -r requirements.txt
```
Install the backend of your choice.
```
$ pip install tensorflow
```
Run tests
```sh
$ pytest tests/test_siamese.py
```
## Development container
To set up the vscode development container follow the instructions at the link provided:
https://github.com/aspamers/vscode-devcontainer
You will also need to install the nvidia docker gpu passthrough layer:
https://github.com/NVIDIA/nvidia-docker
## requirements ## requirements
keras==2.2.4 keras==2.2.4