This project explores the construction and optimization of neural networks for image classification tasks, focusing on the AlexNet architecture. It includes building a basic neural network, optimizing it, implementing and improving AlexNet, and enhancing model performance with data augmentation.
The project covers several key areas:
- Building a Basic Neural Network: Begins with a foundational approach to binary classification, highlighting data preprocessing, model architecture, and performance metrics.
- Optimizing the Neural Network: Investigates the impact of hyperparameter tuning, including dropout rates, activation functions, and optimizers. Techniques such as Early Stopping and K-Fold Cross Validation are employed to enhance model performance.
- Implementing & Improving AlexNet: Adapts AlexNet to classify images into categories such as dogs, vehicles, and food, achieving significant accuracy.
- Optimizing CNN & Data Augmentation: Utilizes the Street View House Numbers (SVHN) dataset, applying data augmentation to improve the model's generalization and performance.
Dataset Link:
https://kaggle.com/datasets/c3a1ad7d0ab948e6bf9f2242ae06247ed5ff8adc894215aa1a292992ea9d99bc
├── datasets/ # Folder containing dataset used in the project
│ ├── dogs/
│ ├── foods/
│ └── vehicles/
Python Code :
AlexNet_Image_Classification_Project/
│
├── basic_NN_construction.ipynb/ # Jupyter notebooks with all coding experiments
├── advanced_NN_optimization_experiments.ipynb/
├── alexnet_implementation.ipynb/
└── alexnet_optimization_and_data_augmentation.ipynb/
- basic_NN_construction.ipynb: Details the construction and initial experimentation of the basic neural network.
- advanced_NN_optimization_experiments.ipynb: Explores further experiments and optimizations.
- alexnet_implementation.ipynb: Demonstrates AlexNet implementations and optimizations.
- alexnet_optimization_and_data_augmentation.ipynb: Advanced optimization and data augmentation (VGG13).
To replicate our findings or build upon them, ensure you have the following:
- Python 3.x
- TensorFlow
- Keras
- NumPy
- Matplotlib for plotting
pip install tensorflow keras numpy matplotlib- Clone this repository.
- Ensure the necessary dependencies are installed.
- Follow the notebooks sequentially to grasp the flow from data preprocessing to model optimization.
The project utilizes diverse datasets, including images of dogs, vehicles, food, and street view house numbers. Key steps in data preparation include:
- Normalization: Pixel values are normalized to have a mean of 0 and standard deviation of 1, facilitating model training and convergence.
- Augmentation: To enhance model robustness, data augmentation techniques like random rotations, resizing, and normalization are applied, creating a more diverse training set.
The initial model is a straightforward neural network designed for binary classification, featuring layers with ReLU and sigmoid activation functions.
The architecture mimics the original AlexNet with adjustments for the specific datasets used. It includes convolutional layers, max-pooling, and fully connected layers, employing ReLU activations and dropout for regularization.
- Early Stopping: Monitors validation loss to halt training preemptively, preventing overfitting.
- K-Fold Cross Validation: Ensures model reliability and generalizability across different data subsets.
- Learning Rate Scheduler: Adjusts the learning rate dynamically, optimizing the training phase.
VGG-13 is a convolutional neural network model known for its simplicity, using 13 layers with weights (including convolutional and fully connected layers) that was developed for large-scale image recognition tasks.
Adapted AlexNet reached a peak test accuracy of 90.18% and a training loss of 0.3100, demonstrating outstanding generalization to unseen data.
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent