MultiLabelCNNClassifier

Introduction

MultiLabelCNNClassifier is a CNN-based framework designed to streamline the process of training custom classifiers. With a straightforward setup and intuitive directory structure, you can easily prepare your image data and start training.

Directory Structure

To get started, place your image data into the Data/Images directory. The expected structure is as follows:

Data
└── Images
    ├── Category
    |   ├── AniName.jpg
    |   ├── AniName.jpg
    |   ├── AniName.jpg
    |   └── AniName.jpg
    └── Category
        ├── AniName.jpg
        ├── AniName.jpg
        ├── AniName.jpg
        └── AniName.jpg

Data Augmentation

After placing your images in the appropriate directory, run the helper.py script within the Data directory using the command:

py helper.py

This script aids in balancing your dataset through data augmentation. Data augmentation is a technique used to artificially increase the size of your dataset by creating modified versions of existing data. This ensures a balanced training process, which is crucial for achieving optimal machine learning model performance.

Model Configuration

If you wish to modify the default settings of the model, navigate to the MODEL SETTINGS section in agent.py. Here, you can adjust parameters such as the training epoch. Additionally, you can configure whether the CNN skips the training phase and proceeds directly to testing.

Model Training

Once you've set up your presets, you can begin training your model by executing agent.py. The system will save a checkpoint in the "Models" directory every 10 epochs. If you wish to retain any of these checkpoints, simply transfer them to a different folder. After doing so, you can utilize the saved model whenever you like. However, ensure that you use the same agent configurations each time. For instance, if you trained your model with 10 classes, you must run it with those same 10 classes.

Model Evaluation

To evaluate the performance of your model, follow these steps:

1. Disable the training mode.
2. Create a folder named "Tests" and place your test images inside.
3. Configure the test section accordingly.
4. Run the evaluation. If set up correctly, the system will process the images from the "Tests" folder and provide predictions based on the trained model. Ensure you review the results to verify the model's accuracy and effectiveness in making predictions.

Visualizing Neural Network Filters

To view the neural network filters, execute the extract-filters script located in the main directory. This will extract and visualize the model's filters. If you've changed the model's name, ensure you update it within this script to correctly track the model.

Customization and Exploration

This framework is designed to be both robust and flexible. If you're experienced, you're encouraged to modify and adapt it to your needs. And even if you're a beginner, don't hesitate to experiment with different settings and values. Through exploration, you'll undoubtedly gain valuable insights and knowledge. Happy coding, and may the force be with you!