Image Augmentation and MNIST Classification with TensorFlow

This project demonstrates image augmentation techniques and trains a neural network on the MNIST dataset using TensorFlow. The goal is to showcase the impact of data augmentation on model performance.

---

Features

• Image Augmentation:

  • Horizontal flipping
  • Grayscale conversion
  • Saturation adjustment
  • Brightness adjustment
  • Rotation adjustment
  • Cropping adjustment

• Dataset Preparation:

  • Preprocessing and normalization of MNIST data.
  • Efficient pipeline with caching, shuffling, batching, and prefetching.

• Deep Learning Model:

  • Neural network with two hidden layers, each having 4096 neurons and ReLU activation.
  • Trained on augmented and non-augmented datasets for performance comparison.

• System Requirements

  1. Python 3.6+

  2. TensorFlow 2.x (Ensure version of Tensorflow compatible with cuDNN library.)

  3. TensorFlow Datasets

  4. Matplotlib

  5. Pillow (PIL)

  6. TensorFlow Docs (optional for visualization) - Ensure version of Tensorflow and cuDNN library.