Title

Exploring the Efficacy of Supervised and Self-Supervised Learning for Wildlife Image Classification: A Case Study in Tiputini, Ecuador

Author

Diego Villacreses
Last Updated: December 1, 2024

Overview

This repository contains the implementation and analysis code for the master's thesis focused on comparing supervised and self-supervised learning approaches for wildlife image classification. The study uses data from the Tiputini Biodiversity Station in Ecuador to evaluate model efficacy in real-world complex computer vision classification problems.

Features

• Implementation of supervised learning models (CNNs, ViT).
• Self-supervised learning using SimCLR.
• Data preprocessing and augmentation pipelines.
• Statistical analysis of model performance.
• Visualization of results (t-SNE plots, graphs, tables).

Repository Structure

.
├── main.ipynb          # Main notebook for experimentation
├── src/                # Source code directory
│   ├── config.py       # Configuration variables
│   ├── utils.py        # Utility functions
│   ├── data_processing # Data loading and preprocessing
│   ├── datasets        # Custom dataset implementations
│   ├── models          # Model architectures and training scripts
│   ├── training        # Training and evaluation functions
├── outputs/            # Directory for saving results (graphs, tables, etc.)

Setup and Installation

1. Clone the repository:

   git clone https://github.com/DiegoDVillacreses/mmia_thesis.git
   cd mmia_thesis

2. Ensure the following directories are set up:
   • data/: Contains labeled and unlabeled datasets.
   • outputs/: Used for saving results like graphs and tables.

Usage

1. Update paths and parameters in src/config.py.
2. Run the notebook:

   jupyter notebook main.ipynb

3. Modify the global variables in the notebook to control experiments:
   • TRAIN_GRID_SUPERVISED: Toggle grid search for supervised training.
   • TRAIN_ENCODER_SIMCLR: Train the encoder using SimCLR.
   • TRAIN_GRID_SIMCLR_CLASSIFIER: Grid search for SimCLR classifier.
   • COMPUTE_STATISTICAL_COMPARISON: Perform statistical tests on model performance.

Results

Results and outputs will be saved in the outputs/ directory. Key outputs include:

• Accuracy and loss metrics.
• Statistical comparison between models.
• Visualizations like t-SNE and classification plots.