tool · draft
DeepReefMap
Open-source pipeline from EPFL's ECEO lab that turns handheld underwater video of coral reefs into labelled 3D point clouds, orthomosaics and benthic cover statistics.
What is reusable: open sourceunknown dataAPI unknownunknown participation Apache 2.0
DeepReefMap is software for rapid 3D semantic mapping of coral reefs from video shot with a handheld camera. It is maintained by Hugues Sibille (EPFL) and Jonathan Sauder (MIT and EPFL) under EPFL's ECEO group, and the method was published in Methods in Ecology and Evolution in 2024. From one video it produces a semantic 3D point cloud, an orthomosaic, per-class benthic cover statistics and an interactive viewer.
The pipeline rectifies frames with a camera profile (GoPro Hero 10 and 12 profiles ship with it; other cameras are calibrated with a built-in COLMAP tool), segments each frame, reconstructs depth and pose, then exports the products. Several reconstruction backends can be chosen, from a basic one that runs on CPU to higher-quality ones that need an NVIDIA or AMD GPU, and one whose weights are gated and limited to non-commercial research use. Segmentation defaults to a model trained on CoralscapesV2, a panoptic reef dataset of 2,433 annotated images from 55 Red Sea sites with 95 classes, released under Apache 2.0.
It is a command-line tool in Python 3.10 to 3.12, licensed Apache 2.0, with a GPU strongly recommended; a separate desktop app exists. Vendored components and downloaded checkpoints carry their own terms.
Status: public and actively developed, with workshop papers cited for 2025 and 2026; no tagged release yet.
- form
- open-source pipeline
- group
- Wildlife and oceans
- homepage
- https://github.com/eceo-epfl/deepreefmap
- licence
- Apache 2.0
- cost
- free
requires this data or input
Densely annotated underwater image datasets of Red Sea coral reefs from EPFL and partners for training segmentation models; V2 adds panoptic labels, 95 classes and fish instances, and is the training data behind DeepReefMap.
open data
uses this engine, platform or technique
Feed-forward 3D vision model from Meta AI and Oxford's Visual Geometry Group that predicts camera poses, depth and point clouds from video; used as a reconstruction backend in DeepReefMap for coral reef mapping.
partly open source
Open-source streaming 3D reconstruction model from the Robbyant team at Ant Group; turns a video or image sequence into camera poses, depth and point clouds in one feed-forward pass.
open source
Tags: coral, reef, 3d reconstruction, segmentation, underwater, video