SpeciesNet
Identify the species caught by a camera trap, among more than 2,000 categories.
SpeciesNet chains two models: MegaDetector finds the animal in the image, then a classifier identifies it, down to the species when it can. This is the approach used for species recognition on the Wildlife Insights platform, according to the project's repository.
And where does AI come in?
An image classifier trained by Google on more than 65 million camera-trap images, geographically diverse. The code is under the Apache 2.0 licence, and the repository explains how to download the model weights and fine-tune them on your own data.
The AI label
- Model or technique
- A detector (MegaDetector) followed by an EfficientNet V2 M classifier covering more than 2,000 labels: species, broader groups, blank image, vehicle.
- Model published
- Not filled in
- Where it runs
- On the user's machine, with a Python package; a GPU helps but is not required.
- Data
- More than 65 million images, including images from the Wildlife Insights community and from public datasets.
- Role of people
- Not filled in
- Frugality
- Not filled in
“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/speciesnet"><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Something wrong or out of date?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.