This repository contains image classification on two common datasets (CIFAR-10 and MNIST datasets) using the LeNet-5
architecture.It includes data loading, model definition, training, evaluation metrics, and visualizations.
-
Data Loading and Exploration:
- CIFAR10 dataset: 60000 32x32 colour images in 10 classes.
- MNIST dataset: Handwritten Digit of flower images.
-
Data Visualization:
- Sample images from CIFAR-10 and MNIST are visualized .
-
Model Architecture:
- Utilizes the
LeNet-5 architecture
to accomplish image classification tasks for CIFAR-10 and MNIST. - Different input shapes are used for RGB images (CIFAR-10) and grayscale images (MNIST).
- Utilizes the
-
Evaluation Metrics and Confusion Matrices:
- confusion matrices are visualized.
- Precision, recall, and F1 score are calculated for both datasets.
-
Analysis of Model Performance:
- Insights into the challenges faced by the LeNet-5 model on CIFAR-10 are provided.
- Factors include data complexity, model architecture, capacity, and potential overfitting.
-
Conclusion:
- It offers a comprehensive overview of data handling, model architecture, training, and evaluation for CIFAR-10 and MNIST datasets.
- It concludes with insights into the potential limitations of the LeNet-5 model on CIFAR-10.