LiteCNN is a Python library designed to simplify the creation, training, and visualization of convolutional neural networks (CNNs). It provides an intuitive interface for deep learning enthusiasts and developers who want to work with CNN models without the complexity often associated with neural network frameworks.
- Straightforward definition of CNN layers with intuitive syntax
- Streamlined training and model evolution capabilities
- Visual representation of model architecture
- 15 pre-configured popular Keras application models ready for immediate use
- Seamless conversion of Keras models to ONNX format
Languages: Python
Libraries: Tensorflow, Matplotlib, Numpy, OpenCv
- Very easy and comfortable syntax
- Full control for developer
- Automatic data preparation and visualization processes
- Compatibility of model: option to convert to onnx type file.
- Presets for popular models
- Exporter and Converter for files with models
- Special Visualizer to display training process
Contributions are always welcome!
See contributing.md for ways to get started.
Please adhere to this project's code of conduct.
from litecnn.core import LiteCNN
from litecnn.visualizer import TrainingVisualizer
import os
from tensorflow.keras.datasets import cifar10
class_names = ['car', 'plane', 'cat', 'dog', 'bird', 'deer', 'horse', 'frog', 'ship', 'truck']
(x_train, y_train), (x_test, y_test) = cifar10.load_data()
my_file = os.path.join(os.path.dirname(__file__), 'car.jpg')
x_train = x_train[:2000]
y_train = y_train[:2000]
x_test = x_test[:400]
y_test = y_test[:400]
x_train = x_train / 255
x_test = x_test / 255
model = LiteCNN()
model.add_conv(32, 3)
model.add_max_pool(2)
model.add_conv(64, 3)
model.add_max_pool(2)
model.add_conv(128, 3)
model.add_max_pool(2)
model.add_flatten()
model.add_dense(10, activation='softmax')
model.compile()
history = model.train(x_train, y_train, x_test, y_test, epochs=5)
prediction = model.predict(my_file)
visualizer = TrainingVisualizer()
visualizer.plot_training(history)