New AI Model SeaSplat Enhances Underwater Images, Aiding Coral Reef Conservation Efforts

Researchers at the Woods Hole Oceanographic Institution (WHOI) and the Massachusetts Institute of Technology (MIT) have developed a groundbreaking AI model called SeaSplat, designed to enhance the quality of underwater images and generate accurate 3D reconstructions. This innovation addresses the two primary challenges of capturing underwater imagery: haze and discoloration, both of which significantly distort the visuals. SeaSplat works by transforming murky, washed-out underwater photos into vibrant, sharp images that accurately represent the true colors and details of objects and organisms such as coral reefs. The model employs a technique known as 3D Gaussian splatting, which allows it to predict and digitally recreate comprehensive 3D scenes from 2D underwater photographs. By considering the 3D structure, geometry, and relative positions of all elements in the scene, SeaSplat can produce images that are nearly identical to how they would appear in air. The development of SeaSplat began with extensive data collection using underwater robots and divers equipped with standard cameras. An important tool in this process was an NVIDIA Jetson Orin edge computing device, which helped guide and control the submersible robots. After the initial training phase—which utilized powerful NVIDIA L40 GPUs—the model could be deployed to process images from regular underwater cameras without requiring specialized equipment or additional lighting. The significance of SeaSplat lies in its potential to aid marine ecologists in monitoring and understanding the complex ecosystems within coral reefs. Coral reefs, often referred to as the "rainforests of the sea," are crucial for biodiversity and support an array of marine life. However, they are under severe threat, with approximately 84% of the world's reefs having experienced harmful bleaching since 1998, according to the International Coral Reef Initiative. The ability to accurately assess the health and structure of coral reefs is essential for effective conservation efforts. Yogesh Girdhar, an associate scientist at WHOI and co-creator of SeaSplat, highlighted the importance of this tool for scientific research. "Scientists can use this model to quantify the biodiversity of reefs and detect specific events like coral bleaching or disease," he stated. Girdhar, along with Daniel Yang, an MIT graduate student, and John Leonard, an MIT professor of mechanical engineering, presented their findings at a robotics conference in mid-May, just ahead of World Reef Awareness Day on June 1. One of the key benefits of SeaSplat is its versatility and the high quality of the 3D models it generates. Ecologists can upload raw underwater images, and SeaSplat processes them to produce corrected versions with restored natural colors and enhanced details. For instance, the model accurately reintroduces reds and yellows, which are typically lost in underwater environments due to water's absorption of certain wavelengths of light. It also highlights fine features of corals and marine organisms, making it easier to identify and study individual elements within the reef. SeaSplat has already been applied to analyze and improve images from various locations, including the U.S. Virgin Islands, the Red Sea, and Curaçao. These successes demonstrate the model's current capabilities and pave the way for future enhancements. The researchers are now working to refine SeaSplat to make it more generalizable and scalable, so it can be used effectively in a broader range of underwater surveys and studies. In addition to its immediate applications, SeaSplat represents a significant step forward in ecological research and conservation. The model provides a powerful tool for ecologists to monitor changes in coral reefs more accurately, contributing to the protection of these critically important marine habitats. As Girdhar noted, the high-resolution 3D models allow for virtual exploration from multiple angles, giving scientists a detailed and realistic view of the sea floor. Industry insiders have praised the innovation, noting that it could revolutionize the way marine biologists and ecologists conduct their research. The combination of advanced AI and practical deployment methods makes SeaSplat a cutting-edge solution for underwater imaging challenges. WHOI, known for its leading role in oceanographic research, and MIT, a renowned institution for technological advancement, bring strong reputations to this project, ensuring that SeaSplat will be rigorously tested and continuously improved to meet the needs of the scientific community.