aemonge / alicia

A CLI to download, create, modify, train, test, predict and compare an image classifiers.

AlicIA

Usage: alicia [OPTIONS] COMMAND [ARGS]...

  A CLI to download, create, modify, train, test, predict and compare an image classifiers.

  Supporting mostly all torch-vision neural networks and datasets.

  This will also identify cute 🐱 or a fierce 🐶, also flowers or what type of
  🏘️ you should be.

Options:
  -v, --verbose
  -g, --gpu
  --version      Show the version and exit.
  --help         Show this message and exit.

Commands:
  compare   Compare the info, accuracy, and step speed two (or more by...
  create    Creates a new model for a given architecture.
  download  Download a MNIST dataset with PyTorch and split it into...
  info      Display information about a model architecture.
  modify    Changes the hyper parameters of a model.
  predict   Predict images using a pre trained model, for a given folder...
  test      Test a pre trained model.
  train     Train a given architecture with a data directory containing a...

View a FashionMNIST demo

Install and usage

pip install alicia
alicia --help

If you just want to see a quick showcase of the tool, download and run showcase.sh https://github.com/aemonge/alicia/raw/main/docs/showcase.sh

Features

To see the full list of features, and option please refer to alicia --help

• Download common torchvision datasets (tested with the following):

  • MNIST
  • FashionMNIST
  • Flowers102
  • EMNIST
  • StanfordCars
  • KMNIST and CIFAR10

• Select different transforms to train.

• Train, test and predict using different custom-made and torch-vision models:

  • SqueezeNet
  • AlexNet
  • MNASNet
  • LetNet5
  • Basic
  • Elemental
  • BasicConv

• Get information about each model.
• Edit/modify a model, changing the features and classifier.
• Compare models training speed, accuracy, and meta information.
• View test prediction results in the console, or with matplotlib.
• Adds the network training history log, to the model. To enhance the info and compare.
• Supports pre-trained models, with weights settings.
• Automatically set the input size based on the image resolution.

• Visualize the model full information include:

  • Model Name.
  • Model pth file location.
  • Model size: memory, disk and state dict.
  • Transforms used: train, valid, test and display.
  • Data directories paths; train, valid and test.
  • Labels of the model.
  • Features of the model.
  • Classifier for the model.

References

Useful links found and used while developing this

• https://medium.com/analytics-vidhya/creating-a-custom-dataset-and-dataloader-in-pytorch-76f210a1df5d
• https://stackoverflow.com/questions/51911749/what-is-the-difference-between-torch-tensor-and-torch-tensor
• https://deepai.org/dataset/mnist
• https://medium.com/fenwicks/tutorial-1-mnist-the-hello-world-of-deep-learning-abd252c47709