ImaginAIry is a custom-built Convolutional Neural Network (CNN) model designed for image classification tasks. This project aims to train a deep learning model from scratch on your own dataset for classifying images into different categories. The model uses CNNs, a popular architecture for image-related tasks.
- Image Classification: Classify images into predefined categories.
- Train from Scratch: No pre-trained models—train your own model.
- Customizable Architecture: Modify the number of layers, neurons, and other parameters.
- Training and Testing Pipelines: Set up a training pipeline to train the model and a separate pipeline to evaluate it on unseen data.
data/: This directory contains the image dataset. Organize it into subdirectories liketrain/andtest/.src/: Source code files for the model.cnn_model.pycontains the model architecture, andutils.pycontains data loading and preprocessing functions.outputs/: Folder to store model outputs like training logs, checkpoints, and predictions.requirements.txt: Contains all the Python dependencies required for the project.main.py: Script that runs the training and evaluation of the model.
- Set up the project structure.
- Build the CNN model architecture from scratch (without using pre-trained models).
- Implement data preprocessing functions (loading and augmentation).
- Train the model on the image dataset.
- Evaluate the model using testing data and visualize performance.
- Implement model saving and loading functions.
- Optimize model performance (tune hyperparameters).
- Create a custom dataset loader for different image formats.
- Experiment with more advanced CNN architectures (e.g., ResNet).
- Implement data augmentation for better generalization.
- Add support for multi-class and multi-label classification.
This project is licensed under the MIT License - see the LICENSE file for details.
