DeepReefMap:
Scalable Semantic 3D Mapping of Coral Reefs with Deep Learning


EPFL Logo
TRSC Logo ECEO Logo

This is the project page of DeepReefMap, an in-development method for semantic 3D mapping of coral reefs from underwater video. DeepReefMap is being developed within the Transnational Red Sea Center at EPFL in collaboration with partnering researchers in the Red Sea.

Towards Fully Automatic Analysis of Video Transects

  • Automation: DeepReefMap takes input only the video, and outputs the 3D reconstruction including benthic class segmentation, alleviating the need for human intervention.
  • Easy data collection: ego-motion video from affordable underwater cameras is the only needed input! A 100m transect can be covered in less than five minutes!
  • Fast: The 3D reconstruction and segmentation runs in real-time on a computer with a consumer GPU!

Video is collected by simply swimming forward while being 1-4 meters above the reef substrate. Then, video frames are extracted from the input video and fed into the 3D reconstruction framework, which estimates the pixel-wise distance to the camera, and tracks the camera movement. The frames are also passsed through a semantic segmentation network. Combing the outputs of these two systems enables the creation of point clouds at 18 frames per second.

Pipeline of DeepReefMap

Robust to Challenging Scenes

Transnational Red Sea Center

The Transnational Red Sea Center is a scientific research center created in 2019 at the Ecole Polytechnique fédérale in Lausanne (EPFL) with the official support of the Swiss Foreign Ministry. An independent and not-for-profit organization, the Center capitalizes on Switzerland’s neutrality, its longstanding tradition of promoting dialogue and its reputation for scientific excellence in order to bridge science and diplomacy for the future of coral reefs. In the Red Sea and beyond.

BibTeX

@misc{sauder2023scalable,
    title={Scalable Semantic 3D Mapping of Coral Reefs with Deep Learning}, 
    author={Jonathan Sauder and Guilhem Banc-Prandi and Anders Meibom and Devis Tuia},
    year={2023},
    eprint={2309.12804},
    archivePrefix={arXiv},
    primaryClass={cs.CV}
}