MegaDetector
Sort camera-trap images, so that conservation teams get their time back.
Camera traps produce millions of images, 70 to 95% of which are empty according to the project: wind, rain, a moving branch. MegaDetector finds and boxes animals, people and vehicles, so that teams only look at the images that matter. More than 80 conservation organisations use it, according to its repository.
And where does AI come in?
An object detection model, released under the MIT licence. It does not identify species: it finds the animal and hands it over to a classifier, which lets it work across very different ecosystems.
The AI label
- Model or technique
- Object detectors (YOLOv9, YOLOv10 and RT-DETR for version 6); the compact variant has 2.3 million parameters.
- Model published
- Not filled in
- Where it runs
- On the user's machine, with the PyTorch-Wildlife Python package; an online demo is also available.
- Data
- Not filled in
- Role of people
- Users set the confidence threshold, usually between 0.15 and 0.3, and stay in charge of the sorting.
- Frugality
- Version 6 aims for smaller, faster models: the compact variant is 2% of the size of version 5.
“Not filled in”: MAUB could not check it with the project. Know more? Write to us.
Run this project? Show the “Listed on MAUB” badge
Paste this code on your site: an image and a link to this page. No script, no tracker: MAUB cannot tell who displays it.
<a href="https://maub.subalpine.it/en/projets/megadetector"><img src="https://maub.subalpine.it/statique/img/badge-recense-en.svg" alt="Listed on MAUB" width="180" height="48"></a>Something wrong or out of date? Write to contact@subalpine.it.
The monthly newsletter
Once a month, what moved on MAUB, in your inbox.