Taiwan Launches First AI Marine Debris Dataset, Initiates International Competition

Taiwan's National Oceanic Research Institute (NORI) has unveiled the country's first AI-ready dataset specifically for marine debris imagery, dubbed 'MDImageNet'. The launch on July 30 marks a significant step in leveraging artificial intelligence for marine environmental protection.

Alongside the dataset release, NORI has also kicked off the '2026 International AI Competition for Marine Debris Image Recognition'. This competition aims to accelerate the digital transformation of marine governance and boost the efficiency of monitoring marine waste. Traditional coastal surveys, which rely heavily on manual collection, classification, measurement, and data entry, are time-consuming and can suffer from inconsistent interpretation standards.

“

The National Oceanic Research Institute has already built an integrated platform for national marine-related data, accumulating over 30 billion data points. In the future, we will continue to promote data standardization, structuring, and governance work, creating more possibilities for Taiwan's marine science and technology development.

NORI has spent years developing the MDImageNet dataset, which currently comprises over 20,000 real coastal images. These images feature more than 42,000 annotated instances of marine debris, creating a high-quality resource tailored for training AI models. The institute has also released a Baseline Model developed by its research team to serve as a starting point for competition participants.

Ocean Affairs Council Vice Minister Wu Hsin-hsiu highlighted NORI's role in building a platform that integrates national marine data, holding over 30 billion records. He emphasized the ongoing commitment to standardizing and structuring this data for future advancements in marine technology. NORI President Chen Chang-ling added that the MDImageNet dataset represents the digitization of years of coastal survey results and will be shared openly to facilitate AI model training and technological development by research institutions, academia, and industry.

“

The release of the MDImageNet marine debris image dataset not only represents the digital organization of years of coastal survey results but will also allow research units, academia, and industry to jointly utilize it for AI model training and technology development through an open sharing mechanism.