The dataset comprises 90 different animal images. Initially, l structured for one-vs-rest classification, followed by binary classification and then a 5-class classification problem. l evaluated each model's performance using classification matrices.
-
Dataset Preparation:
- Organize the dataset for one-vs-rest classification. Perform binary classification using existing architectures and then restructure for 5-class classification. Use 3-fold cross-validation to assess the model.
-
Model Development:
- Build a custom CNN model without using existing architectures like ResNet or DenseNet.
-
Training and Evaluation:
- Train the model on prepared datasets for one-vs-rest and 5-class classification.
- Generate classification matrices for visualization.
-
Convolutional Layer Visualization:
- Plot the output of all convolutional layers and discuss the insights on automatically created features.