Marine Debris Detection Project

This repository contains a Real-Time DEtection TRansformer (RT-DETR)-based project for detecting marine debris on sandy beaches. The dataset (made by the owners of this repository: Nathancgy and KrishManan) includes approximately 400 images categorized into seven types of debris: plastic bottle, styrofoam, plastic container, plastic bag, can, and tire. Each image is annotated with bounding boxes suitable for YOLOv10/RT-DETR training.

Features

• Data: 400 images with annotations for seven categories of marine debris.
• Model: Fine-tuned YOLOv8l, YOLOv10l, and RT-DETR model for accurate prediction of plastic waste.
• Flask Web App: A website to process uploaded videos and display the results.
• Evaluation: Confusion matrix and F1 curve for RT-DETR training.

Validation Label

Validation Prediction

Sample Result in real time detection video.

demo.mp4

Confusion Matrix

F1 Curve

Installation

Clone the repository and install the required packages:

git clone https://github.com/Nathancgy/marine_debris_proj.git
cd marine_debris_proj
pip install -r requirements.txt

Training

Configure the config.yaml file to have the paths to your training images.

config.yaml

Run the train script.

train.py

or use the CLI command

yolo task=detect mode=train epochs=[num of epochs] batch=[num of batches] plots=True model=[your weights] data=config.yaml imgsz=1000

Usage

• Weights: Pretrained weights are not yet released. They are being verified and improved for further usages. Run the detection script with customed weights:

python run.py

or use the CLI command

yolo detect predict model=[Your Weights] source=[source].jpg

Flask Web App

To start the Flask web app for processing uploaded videos:

python app.py

Upload a video and view the processed results on the website.

Future Work

This project is in the development phase and aims to be implemented in real-world scenarios.

Feel free to modify or expand upon this as needed for your project.