OceanEye

The test results of the trained model
Plot results obtained for YOLOv7
batch train

Inspiration

The oceans are facing an environmental crisis caused by tons of trash. As stewards of the planet, it is our responsibility to find innovative solutions to protect marine life and ecosystems. Ocean trash detection presents a unique challenge that requires cutting-edge technology and collaborative efforts. By working together, we can create a sustainable future and ensure that the oceans remain a healthy and thriving environment for generations to come.

Problem statement

The increase in ocean pollution has led to the accumulation of trash and debris in the deep sea, affecting marine life and causing irreversible damage to the ecosystem. There is a need for an automated system that can detect and classify the types of trash present in the deep sea, to enable effective clean-up efforts and prevent further harm to the environment. The aim of this project is to develop a deep learning-based solution to detect and classify ocean trash in the deep sea using underwater images and videos.

What it does

Our YOLOv7-based application detects and classifies ocean trash from real-time images uploaded by users. It identifies trash even in blurry vision, allowing comprehensive analysis of ocean pollution. With our tool, individuals, organizations, and policymakers can take action to create a cleaner, healthier future for our planet's oceans.

How we built it

Our Yolov7 model was trained on the Taco Dataset to improve its accuracy, utilizing PyTorch. To make the model accessible to users, we developed a user-friendly web interface using Chakra UI. Fastapi was utilized as the backend to manage input images and predict object classifications. The resulting output provides a comprehensive object detection and classification system, capable of detecting and categorizing various objects with high accuracy.

Challenges we ran into

The model training process took longer than expected, requiring significant time investment to achieve desired accuracy levels.
Improving model accuracy was challenging, requiring multiple iterations and adjustments to the training process.
Integrating the trained model with an API to enable real-time predictions presented its own set of challenges, requiring careful consideration of both technical and practical requirements.

Accomplishments that we're proud of

Successfully training the YOLOv7 model on the Taco dataset, achieving high accuracy levels and outperforming competitors in the field.
Developing a user-friendly web application that utilizes the trained model to detect and classify ocean trash in real-time, contributing to efforts to reduce plastic pollution and protect marine life.
Establishing a scalable and sustainable platform that can be adapted and expanded in the future to address additional environmental challenges.

What we learned

We gained knowledge and experience in object detection and classification techniques, particularly through the utilization of YOLOv7 for accurate detection.
We also honed our skills in developing a user-friendly web interface using Flask, enabling easy interaction with the model and contributing to the overall success of the project.

What's next for OceanEye

Exploring the feasibility of deploying the model on edge devices to enable real-time object detection and classification in remote areas with limited internet connectivity.
Investigating the possibility of integrating other datasets to improve the accuracy of the model and tuning the hyperparameters for better performance.