iiinova — where your digital self keeps evolving

try it: lab.wildflow.ai 🪸🪸🪸

000

🔵 Please annotate a few colonies. It would help us a lot! Together, we map every reef on Earth! 🤗 🌍

100

And to the amazing people doing the real work in the water, who kindly shared this reef data with us: Rachel Alderdice, Christian R Voolstra, Emma Camp, Paige Strudwick, Christine Roper, and many others.

100

🔵 Huge thanks to the superstar Matthew Osborne, creator of The Living Atlas, for leading the development! The fantastic Jan Paulini built our first species list.

100

🔵 Please break it. This demo is one tiny patch of reef, held together with duct tape and full of bugs. But even in this state, no other browser tool lets a team annotate a 3D reef together like this.

100

🔵 Under the hood. We project DINOv3 features from the photos into the 3D splats. Then we use LDA to adapt them to your categories and make the model lightweight enough to work in the browser.

100

🔵 Semantic search. Start with a colony, coral fragment, or disease patch. Hit the ≈ button and find visually similar things across the reef. Create a category and add the right matches. Keep going until you’ve mapped it. Then move on to the next one.

100

Those errors wreck downstream AI models. Luckily, we have an unfair advantage. Each coral appears in many photos from different angles. Our selector works in a shared 3D embedding space, across every view at once. Just Shift+click, drag out a semantic radius, and select anything directly in 3D.

100

🔵 Smart selection. Corals are painfully hard to annotate! Their shapes and colours are all different, and it's hard to see where one colony ends and another begins. Even amazing 2D models like SAM struggle with reef photos and orthos. Lousy masks and vague point annotations create bad training data.

100

Leave comments at exact locations. Use @ tags for species, health, etc. Pretty much like Google Docs or Figma, but for a 3D reef.

100

🔵 Dive buddies. Annotating hectares of reef is soo lonely. So we built multiplayer! Bring your friends and annotate together in 3D. See where everyone is looking in real time and split up the work.

100

But the data is enormous. Even small survey plots are hundreds of gigabytes. We already process and stream it directly in the browser. Now we turn it into a workspace.

100

🔵 Ground truth is everything. We start with the highest-signal data: time-series 3D. You see each colony from many angles, at true scale, and track how it evolves over time.

100

Of what works, what doesn't, to coordinate precise action. The evidence is buried in a humongous amount of data scattered across organisations, hard drives, and disconnected tools. We need one place to share the data, examine it, analyse it, and turn it into action. That’s why we’re building this!

100

🔵 Why? Coral reefs are such precious, beautiful, incredibly complex, and threatened ecosystems. They are dying fast. A lot of amazing people are fighting to keep them alive. We need to deploy more solutions faster. That requires a deep understanding of how coral reefs function.

100

Sergei Nozdrenkov @nozdrenkov.com · 25/07/2026

🪸 [Call for contributors!] 🪸 New release: "Google Docs for Coral Reefs" is live!! You can now explore the reef with your dive buddies, select coral colonies, tag species, add comments, and search for visually similar colonies across the reef, all in 3D!

173

ICRS 2026 in New Zealand been a blast 🪸🪸🪸

010

🔵 We can be the first generation that leaves nature better than we found it! We can only do this together 🌏 Links: 🔵 huggingface.co/datasets/wil... -- dataset 🔵 wildflow.ai/soneva -- explore the reef (computer works better) 🔵 www.instagram.com/soneva.coral - Soneva CSM 🪸 bsky.app/profile/nozd...

000

🔵 This dataset was made possible by the amazing efforts of many brilliant people: Ben Yuen, Ayla Kroon, Tom Freer, Ashley Wallis, the Soneva CSM coral interns, and Johanna Leonhardt.

100

🔵 Licence CC-BY-4.0 - no strings attached, even for commercial use. Enjoy! 🤗

000

so scientists could track how coral diseases propagate across the ocean, which restoration methods work, etc.

200

3D semantic segmentation across research institutions, AR/VR public engagement, and more. Time-series 3D is even more cool, so we’re super excited to see what comes out of that! My dream is to create an open dataset with thousands of reefs...

100

🔵 What this enables. Last year we open-sourced only 13 plots from Indonesia, and it's already been used all over the world: for underwater video games, improving 3D reconstruction / SLAM algorithms, coral species ID and...

100

Not many regeneration projects have the resources - or the guts - to openly share this level of detail about what they are doing. We hope this bold act triggers a ripple effect across other restoration projects to share their data too!

100

🔵 Unprecedented transparency. All this data comes from the Soneva Conservation & Sustainability Maldives team! They are the ones doing the hard work in the water: restoration, monitoring, and long-term stewardship of these reefs.

100

3 weeks ago we published our 3D time-series viewer, and today it's officially available for all these Soneva sites! You can fly around and track colonies over time right in your browser.

100

🔵 How? We used a high-accuracy algorithm for the co-registration of these time-series surveys. Because of this, you can track how each individual coral colony evolves over time.

100

what methods work in these regions, and allow global access to these discoveries for meaningful, coordinated action. Democratising access to ocean data is key. That's why we are doing it.

100

A lot of amazing people are fighting against this. To win this battle, we need to deploy more accessible solutions faster. We need to understand and share how coral reefs function in different regions...

100

🔵 Why? All coral reefs in the world combined are only roughly the size of New Zealand. Yet they support over 1/4 of all marine life and up to a billion people! Coral reefs are such precious, beautiful, incredibly complex and threatened ecosystems. They are disappearing fast.

100

Right now it’s nearly impossible to find high quality open time-series 3D photogrammetry data of coral reefs. So we’re trying to move the needle. We’re sharing raw images, co-registered 3D models (camera poses, point clouds, splats), colour correction filters, everything.

100

Partnering with a Maldivian NGO Soneva CSM, we’re open-sourcing 23 Coral Reef survey plots: 6 locations, 22k pics (67GB) under CC-BY-4.0 🌏🪸 Time-series 3D photogrammetry data to accelerate research/conservation. No strings attached 🤗 huggingface.co/datasets/wil...

140

We are getting closer to digitizing the ocean at scale. What a time to be alive! 🪸�

010

Huge props to Josh for the execution, and his supervisors Jeff Clark (@savvyscientist.bsky.social) and Rob Jones for guiding this. Play with the live demo in the browser here (drag the slider!): joshhyde04.github.io/Semantic-Spl...

100

🔵 Next Steps: Josh is looking to quantitatively validate these clusters. If you have a "Gold Standard" fully annotated benthic dataset and want to help push this forward, reach out to Josh!

100

Even if generic DINOv3 (trained on internet images, although it’s trained on iNaturalist) isn't perfect for underwater yet, the pipeline is here! We can swap it for an Ocean-Foundation model later, and it will just work.

100

Long term: "Semantic Search" for the ocean. Imagine typing "Show me all branching colonies on this reef" or "Highlight the bleached areas," and the 3D digital twin just lights up.

110

🔵 What does this enable? Short term: much faster 3D segmentation. Instead of painting pixels, we can just cluster these features in 3D.

100

2. Extracts the feature embeddings (the mathematical "meaning" of the texture/shape). 3. Projects those features directly into the 3DGS scene (Semantic Splatting). This machine “sees” the difference between structures and textures automatically.

100

🔵 How does this work? This is Zero-Shot learning. The model wasn't explicitly trained on "corals" or "sand." Josh built a pipeline utilizing DINOv3 (Vision Transformers). 1. He takes raw underwater images (from the open dataset we published earlier).

100

Right now, that often means humans manually tracing pixels for hours (segmentation). It’s painfully slow, expensive, and doesn't scale. We need to interpret entire reef-scapes instantly to coordinate precise action.

100

🔵 Why? Coral reefs are so precious, beautiful, incredibly complex and threatened ecosystems. They are dying fast. Over half a billion people and a quarter of ocean life directly rely on them.To protect them faster, we need to understand corals - measure growth, detect diseases, map species.

100

Fusing DINOv3 Embeddings into 3D Gaussian Splatting of Coral Reefs – check out this incredible work by Josh Hyde from the University of Bristol! 🤯 joshhyde04.github.io/Semantic-Spl... -- demo @savvyscientist.bsky.social

120

wildflow.ai/apply - link to apply wildflow.ai/roadmap - roadmap wildflow.ai/vision - vision

000

🔵 If you know a superstar like that, please DM me! Or comment. Form to apply is in the comment below. 🔵 Billions of people depend on our ability to protect natural ecosystems. There are no laws of physics stopping us. We can do it together! 🌎🪸

100

🔵 We’re bootstrapping our way to product-market fit, collaborating with mission-aligned partners, and exploring a potential raise at some point within a year. Expect early-stage scrappiness with long-term upside (equity, impact, visibility).

110

- Relentless optimist, full of humility and compassion - this work is incredibly hard and full of setbacks, but so worth it - Experience with large-scale deep learning, vision-language models (e.g., CLIP, Segment Anything), 3D/point cloud processing, or ecosystem simulation methods is a big plus.

110

- Excited to work in small, dynamic teams that achieve big results - Ready to travel to Madagascar, Maldives, Indonesia, or French Polynesia, etc - Based (or willing to be) in London or SF/LA - our HQ is still TBD

100