Presented by OceanX's TIDE Initiative, OceanX invites graduate and post-doctoral researchers, and emerging professionals to participate in the OceanX Science Impact Challenge. This multi-phase initiative aims to harness innovative data science and AI expertise to advance OceanX’s AI R&D program, including our AI platform and digital twin projects.
From October 2025 through March 2026, participants will work remotely in teams (2-3 people) to develop robust data solutions and prototypes that address critical marine science challenges, such as enhancing geospatial understanding, visualisation of abstract data and streamline data analysis for mapping. They will leverage real-world multimodal data collected in the field, including:
High-resolution mapping data (i.e. bathymetry, seafloor imagery)
eDNA samples (i.e. geographically correlated genetic information )
ROV and submersible footage (i.e. geo-referenced visual observations of marine life and habitats)
Environmental sensor data (e.g. nutrient measurements, temperature, salinity, depth)
Participants will hone critical skills, build a valuable portfolio project, and network with fellow innovators and OceanX experts.
Participants will directly contribute to the groundbreaking OceanX’s Research & Development program and make a real-world impact on ocean management and conservation.
Finalists will present their work in person and gain unparalleled access to OceanX leadership and marine science experts.
The winning team will be invited to board the OceanXplorer science vessel to see their solution implemented and tested in a real-world ocean environment.
As part of the challenge, teams will be asked to propose questions they believe are key to advancing ocean science and exploration. To spark ideas, we’ve shared a few example questions below. Participants are welcome to use these as inspiration, or suggest entirely new prompts of their own when applying.
At OceanX, we develop AI Models to analyze underwater footage and classify vulnerable marine species to protect, such as corals and sponges. Besides our own, there already exist a variety of models developed for the same tasks, such as the Megalodon (for any fish in the water column).
The problem with using these models is that they greatly underperform when applied to new locations.
Nevertheless, recent AI techniques have been developed to adapt these models to new locations, keeping a similar high accuracy. These techniques are called “Domain Adaptations” and some examples are ADAPT, Optimal Transport, and Align and Distil.
The problem statement for this project is to test techniques for adapting these systems to our data, with the goal of outperforming benchmark accuracy.
Currently, we achieve this by using the “Segment Anything Model” from Meta (specifically, the AutomaticMaskGenerator of SAMv2). This model creates masks (which we convert in bounding boxes) around any object it deems interesting.
The problem statement for this project is to develop a similar model that overcomes these challenges and outperforms our baseline SAM model at detecting is that this model fails to detect marine life with more nuanced shapes, such as whip corals or elongated fish, like eels. One more challenge is that our SAM model has a high false positive rate, meaning that it missclassifies many background objects as “interesting” (i.e. rocks, sandbars, full frame, part of the vehicle).
The problem statement for this project is to develop a similar model that overcomes these challenges and outperforms our baseline SAM model at detecting Marine Life.
The vastness and inaccessibility of the ocean floor pose significant challenges to comprehensively mapping and understanding marine biodiversity. There is a critical need for advanced predictive models that can leverage diverse oceanographic data to infer the distribution of marine life, particularly in areas where direct observation is limited.
The problem statement for this project is to develop an advanced predictive model that leverages diverse oceanographic data (such as bathymetry, temperature, salinity, oxygen, currents, echosounder and eDNA) to infer the distribution of marine life, with the goal of enhancing our ability to explore, monitor, and protect vulnerable deep-sea ecosystems more effectively.
The vast array of advanced sensing equipment used in OceanX missions generates incredibly rich and diverse datasets. However, the lack of interconnectivity between these disparate data sources means that researchers often spend dozens of hours post-mission manually unifying and reformatting data into a usable state for analysis. This time-consuming process significantly delays insights and reduces research efficiency.
The problem statement for this project is to develop innovative solutions that accelerate the process of unifying diverse OceanX mission data, thereby minimizing the manual effort required by researchers and expediting the pathway from data collection to scientific discovery.