cmaeti/mnist-cnn

MNIST CNN Classifier

• Adapted from Pytorch Examples

Model Overview

• Model Name: MNIST CNN Classifier
• Model Type: Convolutional Neural Network (CNN)
• Dataset: MNIST (Modified National Institute of Standards and Technology)
  • A dataset of 70,000 handwritten digits (28x28 grayscale images) of digits 0-9.
• Task: Image classification (Digit recognition)
• Framework: PyTorch
• Preprocessing:
  • Images are normalized and transformed before feeding into the model (e.g., scaling pixel values to a [0, 1] range).

Model Architecture

• Input Layer: 28x28 grayscale images.
• Convolutional Layers:
  • Conv1: 32 filters of size 3x3, applied to the input image (1 channel).
  • Conv2: 64 filters of size 3x3, applied to the output of Conv1.
• Activation Functions:
  • ReLU activations after each convolutional layer.
• Pooling Layer:
  • Max pooling with a 2x2 window after Conv2.
• Dropout:
  • Dropout rate of 25% after the first convolutional layer and 50% after the fully connected layer to prevent overfitting.
• Fully Connected Layers:
  • fc1: A fully connected layer with 128 neurons.
  • fc2: A fully connected layer with 10 output units (one for each digit).
• Output:
  • Log-softmax activation applied to the final layer to produce log-probabilities for the 10 classes.

Training Details

• Batch Size: (60000)
• Epochs: (14)
• Learning Rate: (1.0 Adadelta)
• Hardware Used: (Google Colab T4 GPU)

Performance Metrics

• Training Accuracy: (99%)
• Training Loss: (0.0273)

Usage Instructions

• How to Use:
  • Clone the repo
  • Dependencies:
  • pip install transformers torch torchvision==0.20.0
  • python app.py

Model Limitations

• This model may struggle with noise or distorted images and is trained on relatively clean data.
• The model is limited to the MNIST digit set and may not perform well on other datasets or noisy, out-of-distribution data.