This model was trained by the Deepfaune initiative. The official location of the model files is the Deepfaune GitLab repository.
These files are copied to this Hugging face repo because of the easy integration with AddaxAI.
Developer
The Deepfaune initiative
Model version
v1.5
Description
The Deepfaune initiative aims at developing 'artificial intelligence' models to automatically classify species in images and videos collected using camera-traps. The initiative is led by a core academic team from the French 'Centre National de la Recherche Scientifique' (CNRS), in collaboration with more than 100 European partners involved in wildlife research, conservation and management. The Deepfaune models can be run through a custom software freely available on the website, or through other software packages or platforms like AddaxAI. New versions of the model are published regularly, increasing classification accuracy or adding new species to the list of species that can be recognized. More information is available at: https://www.deepfaune.cnrs.fr. Documentation: https://deepfaune.pages.math.cnrs.fr/software/.
Notes on this copy
The weights file is the one published at https://pbil.univ-lyon1.fr/software/download/deepfaune/v1.5/, byte for byte. Nothing was stripped or re-saved.
The inference.py here is a port of the author's own classifTools.py to the AddaxAI model interface. It reproduces that file's output exactly. Two details from it are load-bearing and easy to lose:
- the backbone is built at 224x224, not the DINOv3 ViT-L default of 256x256
global_poolmust be"token", because timm's DINOv3 defaults to"avg"and the weights load cleanly either way
The bird head published alongside these weights is not included, for now. It works per crop, on the same class-token embedding: when the main head predicts bird, the embedding goes through an MLP that splits it into eight groups. That maps onto AddaxAI's one-distribution-per-crop contract without trouble. It is left out only because adding it later means a new model id anyway, and the Deepfaune team have more heads coming (small mustelids, sex, age), so it is better done once for all of them.
Classes
- bison
- badger
- ibex
- beaver
- red deer
- golden jackal
- chamois
- cat
- goat
- roe deer
- dog
- raccoon dog
- fallow deer
- squirrel
- moose
- equid
- genet
- wolverine
- hedgehog
- lagomorph
- wolf
- otter
- lynx
- mongoose
- marmot
- micromammal
- mouflon
- sheep
- mustelid
- bird
- bear
- porcupine
- nutria
- muskrat
- raccoon
- fox
- arctic fox
- reindeer
- wild boar
- cow
Links