You're pledging to donate if the project hits its minimum goal and gets approved. If not, your funds will be returned.
All existing AI species detection models are closed set which declares that they only can detect species only within their trained datasets so if it finds something new it either skips it or mis identifies. So far existing systems work with separate visual and acoustic sensors and they are not functioning when one fails. A 2025 review in Machine Learning for Bioacoustics proves that multimodal fusion for field detection is an open direction. The present from DeepSeaNet results prove that only 43.4 percent accuracy for finding new species. This project will evaluate a true benchmark for existing models, how they fails, and what are their weaknesses.
This project will answer can a joint model identify a new species that they never had in training and does this differ when it comes to different environments and would it need to be re trained.
Project fund will used to model training and benchmark, personal runway, and dataset access fees and for conference submission fees.
Two people in team: Sithum Vikasitha Jayasinghe and Chethiya De Silva. We co-founded PartnerLabz a self funded software company which now has international clients and PathFinder which is an AI product where Chethiya built the core AI engine himself. We are both close to graduating in 2026 from Birmingham City University, BSc Software Engineering, studied remotely.
Currently we requested Plymouth Marine Lab for data access and If we don't get a reply, we will move to FathomNet, which has a public dataset, so the project can still proceed. A bigger risk is if our benchmark shows the current models don't fail the way we expect the result would still be valid and useful, just different from our hypothesis and would report it honestly rather than force the data to fit.
Started raising money recently so no initial money raised yet.
There are no bids on this project.