Lang Chai King
May 31, 2025 16:36
Researchers have developed Seasplat, an AI model that converts underwater images into clear 3D scenes, to support ecological scholars who study coral reefs and biodiversity.
Researchers at Woods Hole Oceanographic Institution (who) and Massachusetts Institute of Technology (MIT) introduced an innovative AI model called Seasplat. This model is designed to convert an unclear underwater image to a vivid and very detailed 3D scene, providing ecological scholars with more accurate tools to study and understand underwater environments including coral reefs.
Solving underwater image distortion
Seasplat was developed to cope with the general distortion found in underwater photos, mainly in fog and discoloration. Such distortions often obscure the true colors and details of the underwater scenes. By reconstructing how these scenes will look without water interference, Seasplat can create a sharp image that accurately reflects the natural colors of underwater objects and marine life.
Innovative 3D reconstruction technology
This model allows you to create an accurate 360 -degree image in a two -dimensional photo using a technology known as 3D Gaussian Splatting. The researchers use a vast underwater image captured by both divers and robots, indicating the true life of the underwater scene by improving the model to automatically modify the unique water distortion.
Effect on coral reef research
Seasplat has significant potential for understanding ecological research, especially coral reefs that are important for marine biodiversity. According to the International Coral reef initiative, about 84%of the world’s reefs have been suffering from harmful bleaching over the past two years. This model can help to monitor this ecosystem by quantifying biodiversity and detecting events such as coral bleach or disease.
Use of advanced technology
The development of Seasplat was supported by Advanced Technology, including NVIDIA Jetson Orin Edge Compute for capturing underwater images and NVIDIA L40 GPUs for initial training of models. This model can now be applied to the image of the standard underwater camera, so you can access an extensive ecological application.
Future application
The researchers already used to improve the images of Virgin Island, the Red Sea and Curaçao, USA, plan to make Seasplat more generalized and expanded for various underwater studies. This development promises ecological scholars to provide enhanced tools to evaluate and preserve marine biodiversity.
For more information, you can see the original article on the NVIDIA Developer blog.
Image Source: Shutter Stock