This repository contains code implementations and exercises related to deep learning concepts and projects. It consists of two main folders:
This folder contains practical exercises focusing on various aspects of deep learning. Each exercise is designed to provide hands-on experience with different deep learning techniques and frameworks.
The project folder hosts a specific deep learning project, titled "Cassava Classification." This project aims to classify images of cassava leaves into different disease categories using deep learning techniques. The project utilizes a convolutional neural network (CNN) architecture implemented in PyTorch.
The "Cassava Classification" project involves the following key components:
- Utilization of a pre-trained ResNet18 backbone for feature extraction.
- Implementation of data preprocessing techniques using the Albumentations library for image augmentation and normalization.
- Training of the model using the Adam optimizer and minimizing the Dense Cross Entropy loss function.
- Evaluation of the model's performance using stratified k-fold cross-validation and accuracy score metrics.
- Visualization of the training history using Plotly to monitor model performance over epochs.
- Incorporation of model interpretability techniques such as SHAP and LIME to provide insights into the model's predictions.
- Generation of attribution maps to highlight important regions of input images contributing to the model's decision-making process.
This project demonstrates the practical application of deep learning in solving real-world classification problems in agriculture, with an emphasis on model interpretability and performance evaluation.