This repository contains code for a deep learning model designed for image classification tasks. The model includes a custom softmax function and is implemented using TensorFlow and Keras. The project provides a comprehensive demonstration of creating, training, and evaluating a neural network for image recognition.
- Neural Network Architecture
- Softmax Function
- Getting Started
- Results and Visualizations
- Dependencies
The neural network architecture is designed for image classification tasks. It consists of the following layers:
- Input Layer: 400 neurons (assuming 20x20 pixel images)
- Hidden Layer 1: 25 neurons with ReLU activation
- Hidden Layer 2: 15 neurons with ReLU activation
- Output Layer: 10 neurons with linear activation
The model is compiled using the sparse categorical cross-entropy loss and the Adam optimizer.
The repository includes a custom softmax function (my_softmax) implemented in both NumPy and TensorFlow. This function converts a vector of values to a probability distribution and is used in the output layer of the neural network.
Before running the code, ensure you have the following dependencies installed:
- NumPy
- TensorFlow
- Matplotlib
Clone the repository:
git clone https://github.com/your-username/deep-learning-image-classification.git
cd deep-learning-image-classificationThe model training results, including accuracy and loss curves, can be found in the results directory. Additionally, random images from the dataset are visualized alongside the model's predictions to provide insights into its performance.
- NumPy
- TensorFlow
- Matplotlib
Install dependencies using the provided requirements.txt file:
pip install -r requirements.txt