Camouflaged Object Detection System

This project is an AI-powered system designed to detect camouflaged animals in images and videos. It integrates object detection using a fine-tuned Segment Anything Model (SAM2), species classification, and video frame analysis. The application is aimed at wildlife researchers, conservationists, and technology enthusiasts interested in advancements in AI and computer vision.

Features

• Image Detection: Detect camouflaged animals in still images using advanced segmentation models.
• Species Classification: Classify the detected animals into various species.
• Informational Output: Provide detailed information about the detected animal's habitat and species characteristics.
• Interactive Visualization: Display the detection results with segmentation masks, classification labels, and any other relevant information.

Technology Stack

• Backend Framework: Dash (based on Flask)
• Machine Learning Model: BiRefNet for segmentation and CNN-based models for classification.
• Data Processing: NumPy, OpenCV for video and image processing.
• Visualization: Plotly, integrated with Dash for displaying results, including image overlays and detection boundaries.
• Map API: MapBox https://docs.mapbox.com/help/getting-started/access-tokens/ (note you will need to create an account to generate a token, it is free)

Directory Structure

## Directory Structure

```bash
├── run.py                      # Main entry point for the Dash web app
├── configs/                    # Configuration files
│   └── Species_Labels_1.csv    # CSV file with species labels
├── dashboard/                  # Dash application files
│   ├── __init__.py             # Initialization file for the Dash app
│   └── layout.py               # Layout definition for the Dash app
├── data/                       # Data files and scripts
│   ├── bootstrap_camo.sh       # Script to bootstrap camo data
│   ├── bootstrap_cod10k.sh     # Script to bootstrap COD10K data
│   ├── README.md               # README for data directory
│   ├── resize_images.py        # Script to resize images
│   ├── sample_files/           # Sample files for testing
│   ├── test/                   # Test dataset
│   └── train/                  # Training dataset
├── Dockerfile                  # Docker configuration file
├── fine_tuned_model.pth        # Fine-tuned model file
├── logs/                       # Logs directory
│   ├── classification_resnet_metrics_Validation.csv  # Validation metrics for classification
│   ├── classification_resnet_metrics.json            # JSON file with classification metrics
│   └── segmentation_model_finetune.csv               # CSV file with segmentation model finetune metrics
├── models/                     # Model files
│   ├── seg_model.py            # Segmentation model script
│   └── test_seg.py             # Script to test segmentation model
├── README.md                   # Project overview and setup instructions
├── requirements.txt            # Dependencies and libraries
├── sam2/                       # SAM2 model files
│   └── demo/                   # Demo files for SAM2 model
├── temp/                       # Temporary files
│   ├── config.json             # Configuration file
│   └── predicted_label.txt     # Predicted label file
└── utils/                      # Utility functions
    ├── classification.py       # Classification logic
    ├── detection.py            # Image detection logic
    ├── video_processing.py     # Scripts for extracting frames from videos
    └── transform_data.py       # Data transformation script

Setup Instructions

Installation

1. Clone the repository:

   git clone https://github.com/your-repo/camouflaged-object-detection.git
   cd camouflaged-object-detection

2. Create a file named ".env" in the root directory of the repo. In the .env file, add the following line:

   MAPBOX_ACCESS_TOKEN=<YOUR_MAPBOX_TOKEN_HERE> (with no quotes)

   It should look like this:

   MAPBOX_ACCESS_TOKEN=pk.124135435...

3. Install the required libraries:

   pip install -r requirements.txt

4. Start the Dash web server:

   python app.py

5. Open the app by navigating to http://localhost:8080/ in your browser.

Docker Setup

1. Clone the repository:

   git clone https://github.com/your-repo/camouflaged-object-detection.git
   cd camouflaged-object-detection

2. Create a file named ".env" in the root directory of the repo. In the .env file, add the following line:

   MAPBOX_ACCESS_TOKEN=<YOUR_MAPBOX_TOKEN_HERE> (with no quotes)

   It should look like this:

   MAPBOX_ACCESS_TOKEN=pk.124135435...

3. Build the Docker image:

   docker build -t camouflaged-object-detection .

4. Run the Docker container:

    docker run -p 8080:8080 -v ${PWD}/:/app --gpus all camouflaged-object-detection

5. Open the app by navigating to http://localhost:8080/ in your browser.

Bonus Features

• Improve Real-Time Performance: Optimize the model for real-time detection in live video feeds.
• Expand Species Database: Incorporate more species to improve classification accuracy.
• Interactive Visualizations: Enable more detailed interactive graphs and information displays for the users.