Skip to content

Latest commit

 

History

History
63 lines (39 loc) · 2.55 KB

README.md

File metadata and controls

63 lines (39 loc) · 2.55 KB
Raw

SVHN Classification Using CNN

This repository contains code for training a Convolutional Neural Network (CNN) model to classify images from the SVHN dataset, a collection of street-view house numbers. The model is built and trained using TensorFlow and Keras.

Dataset

  • Train Dataset: train_32x32.mat contains 32x32 color images of digits (from 0 to 9).
  • Test Dataset: test_32x32.mat contains similar images for testing.
  • Each image has three channels (RGB), and the labels represent the digit class (0-9).

Steps Involved

1. Loading the Dataset

The dataset is loaded using the scipy.io.loadmat function to import .mat files. The images and labels are preprocessed:

  • Transposed to have dimensions (number of images, height, width, channels).
  • Labels are one-hot encoded for classification.

2. Data Normalization

All pixel values are scaled to the range [0, 1] by dividing by 255.0.

3. Data Augmentation

To prevent overfitting, several augmentation techniques are applied to the training dataset:

  • Random horizontal flip
  • Random brightness adjustment
  • Random saturation adjustment

These augmentations are applied during training to artificially expand the dataset.

4. Splitting the Dataset

The training data is split into training and validation sets using an 80-20 split. The test dataset remains unchanged for model evaluation.

5. Building the CNN Model

A Convolutional Neural Network is built using TensorFlow’s Sequential API. The architecture consists of:

  • 3 Convolutional layers with ReLU activation and MaxPooling.
  • A Flatten layer to prepare for fully connected layers.
  • A Dense layer with 128 neurons and ReLU activation.
  • A Dropout layer for regularization (0.5 dropout rate).
  • A final Dense layer with 10 neurons for classification (softmax activation).

6. Model Compilation

The model is compiled with the following settings:

  • Optimizer: Adam
  • Loss: Categorical Crossentropy
  • Metrics: Accuracy

7. Training the Model

The model is trained on the augmented training dataset for 20 epochs. During training, validation data is used to monitor performance and avoid overfitting.

8. Model Evaluation

After training, the model is evaluated on the test dataset (which was not augmented). The test accuracy and loss are computed.

9. Plotting the Confusion Matrix

To visualize the performance of the model, a confusion matrix is plotted. This matrix shows the correct and incorrect predictions for each digit class.