๐ ๐๐ถ๐๐ถ๐ป๐ด ๐ถ๐ป๐๐ผ ๐ฎ๐ฌ๐ฎ๐ฑ: ๐๐ ๐ฝ๐น๐ผ๐ฟ๐ถ๐ป๐ด ๐จ๐ป๐ฑ๐ฒ๐ฟ๐๐ฎ๐๐ฒ๐ฟ ๐๐ถ๐ผ๐ฑ๐ถ๐๐ฒ๐ฟ๐๐ถ๐๐ ๐๐ต๐ฎ๐น๐น๐ฒ๐ป๐ด๐ฒ๐ ๐
As my first post of the year, Iโm focusing on what lies beneath the surface - our oceans, where biodiversity faces critical challenges but receives far less attention than terrestrial ecosystems.
On land, we have hundreds of trail camera brands, extensive research, and powerful AI tools like MegaDetector for analyzing animals. Underwater, resources are scarce.
Thankfully, companies like Anemo Robotics are working to bridge the gap by developing solutions to capture and analyze underwater images and videos.
But what about AI?
While terrestrial AI models are advanced and plentiful, the same cannot be said for underwater detection. A standout example is the MegaFishDetector - https://lnkd.in/d77N9NTw which is an open-source, binary fish detector under the MIT license, free for anyone to use.
๐ ๏ธ Thoughts on MegaFishDetector:
1๏ธโฃ Likely the best open-source binary fish detector available.
2๏ธโฃ Limited to fish; it doesnโt detect crabs, starfish, seals, etc.
3๏ธโฃ Detection performance is modest on the dataset i tested on (40-60% at 0.2 confidence).
4๏ธโฃ Lower confidence settings increase detections (true and false), making it useful with tools like Animal Detect for further sorting - talk to us to see what we can do.
Iโve included an example image using MegaFishDetector with an image shared by Anemo Robotics. As you can see, itโs far from perfect, where Anemo Robotics own models are both quicker and more accurate, with the confidence level selected for this demonstration,ย on fishes around the Danish coasts.
