Metadata Augmented Deep Neural Networks for Wild Animal Classification

This repository contains code for classifying camera trap images using various deep learning models and metadata fusion techniques.

Table of Contents

1. Installation
2. Usage and Dataset
3. File Descriptions
4. Model Weights
5. License
6. Contributing

Installation

1. Clone the repository:

   git clone https://github.com/ammarlodhi255/metadata-augmented-neural-networks-for-wild-animal-classification.git
   

   cd metadata-augmented-neural-networks-for-wild-animal-classification
   

2. Create and activate a virtual environment:

   python -m venv venv
   source venv/bin/activate  # On Windows use `venv\Scripts\activate`
   

3. Install the required packages:

   pip install -r requirements.txt
   

   Note: If you're using GPU acceleration, ensure you have the appropriate CUDA toolkit installed.

Usage and Dataset

1. Preparing the dataset:

   • The dataset is collected from NINA Viltkamera dataset. It contains 170,000 sample images along with metadata from the NINA website. The webscrapping code is located inside of the file webscrapper.js. Following is the NINA web interface.

   

   • The dataset is highly imbalanced and many classes contained no images. Our curated dataset can be accessed from two links. The first link contains images_part_1.zip file and the second contains rest of images in a file named images_part_2.zip.
   • Ensure your camera trap images are organized in the expected directory structure. Download the important json files from JSON Files. Each entry in the json "annotations" is a single image file. To get the associated image, append the "filename" of the json entry with the path to the image files. So say your images are stored at C:\NINA\images, and you have a filename "96624a46-4bc6-46c9-9770-370bf6d84718.jpg", then the correct image would be C:\NINA\images\96624a46-4bc6-46c9-9770-370bf6d84718.jpg.
   • Update the bp variable in the scripts to point to your dataset location.
   • Image variety and challenges are shown in the following pictures.

   

2. Run the main classification script:

   python viltkamera_classifier.py
   

3. For metadata-only experiments:

   python metadata_classifier.py
   

4. To evaluate metadata models:

   python evaluate_metadata_models.py
   

Modify the scripts as needed to adjust model configurations, data preprocessing steps, or evaluation metrics.

Model Weights

All the model weights can be downloaded from the URL model-weights. You can load the weights and directly perform classification and inference on animal species.

Source Files Descriptions

• viltkamera_classifier.py: Main script for training and evaluating image classifiers.
• metadata_classifier.py: Implements a metadata-only classifier for experiments.
• fusion_model_v2.py: Defines fusion models combining image and metadata features.
• conventional_models.py: Implements various deep learning architectures.
• evaluate_metadata_models.py: Analyzes results from metadata-only experiments.
• Additional code related to UMAP, extraction of scene attributes, and other processes described in the paper can be found in src and notebooks folders.

License

This project is licensed under the MIT License - see the LICENSE file for details.

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

1. Fork the repository
2. Create your feature branch (git checkout -b feature/AmazingFeature)
3. Commit your changes (git commit -m 'Add some AmazingFeature')
4. Push to the branch (git push origin feature/AmazingFeature)
5. Open a Pull Request