imageGenie 0.0.7

ImageGenie

• Training a model to classify images between different classes .
• This single package lets us harness the power of the state of the art models without any hassle of coding them ourselves.
• Just 3 lines of code and we're done.

Installation

pip install imageGenie

COLAB Notebook Demo

https://colab.research.google.com/drive/1DGgrENv-XTVeRz7PsOm0tpofJFZWn6PU?usp=sharing

Usage

1. Fully Automated Mode

• Folder Structure

main folder

from imageGenie.classify import Classifier # import the Classifier Class

cl = Classifier("/root", "/models") # arg1 -> base directory containing train & test ; arg2 -> saving directory

cl.run() # this trains the model by automatically finding out number of classes, types of images and optimum training epochs.

2. Controlled mode (Work in progress)

TODO

• Handle all image formats
• Parse the specifications provided by the uer from a config file. That may include the priority of speed, accuracy, emphasis on False Positives or negatives, time available to experiment and train.
• Include all other model architectures like EfficientNet, MobileNet, Inception, VGG.
• Algorithm to figure out what architecture and hyper-params would be the best (in the fully automated mode) as per hardware.
• Save all other artefacts like pipeline, metrics, plots, etc
• Allow user to construct a model by themselves
• Allow to either have a proper folder structure or a json with labels.