Model zoo

Every model AddaxAI ships, in one table. Select a model to read its full description, licence and citation.

Which one should you pick

A run uses one detection model, and optionally one classification model and one embedding model. They do different jobs.

Detection finds animals, people and vehicles and draws boxes around them. It runs first, on every file. MegaDetector v5a is the default and works almost everywhere, so leave it alone unless you have a good reason and know what you are doing.
Classification names the species inside each animal box. SpeciesNet covers the whole world and is the safe starting point. The regional models know fewer species but were trained on one ecosystem, so they generally do a better job where they apply.
Embedding turns each crop into a fingerprint, which is what powers the similarity sort and the suggestions on the Labels page. It changes no labels of its own.

Looking for one particular animal? Type it into the search box. It looks inside the species lists, so searching "wolverine" leaves only the models that can name one. Select a model to see everything it knows.

44 models, select one to read more

Model	Type	Developer	Region	Summary
`SPECIESNET-v4-0-2-A`#	Classification	Google	Global	Global • 2,000+ classes • Google Research
`EUR-DF-v1-3`#	Classification	The DeepFaune initiative	Europe	34 classes • February 2025 • added bison, moose, reindeer, wolverine
`SAH-DRY-ADS-v1`#	Classification	Addax Data Science	Africa	East & Southern Africa savannas • 328 classes • Trained on 2.8M images
`TERRAI-NEP-v1`#	Classification	Alexander Merdian-Tarko	Asia	10 classes • Tiger conservation
`TAS-BB-v1`#	Classification	Barry Brook	Oceania	96 classes • Trained on 2.5M images
`NAM-ADS-v1`#	Classification	Addax Data Science	Africa	30 classes • Trained on 850k images
`NZI-ADS-v1`#	Classification	Addax Data Science	Oceania	Best on rodents • 17 classes • DOC New Zealand
`PAM-SDZWA-v1`#	Classification	San Diego Zoo Wildlife Alliance	Americas	53 classes • San Diego Zoo Wildlife Alliance
`TKM-ADS-v1`#	Classification	Addax Data Science	Asia	Southern region • 14 classes • Trained on 1M images
`KIR-HEX-v1`#	Classification	Hex Data	Asia	Manas v1 • OSI-Panthera • Trained on 42k images
`SWUSA-SDZWA-v3`#	Classification	San Diego Zoo Wildlife Alliance	Americas	Version 3 • 27 species • Trained on 92k images
`WUSA-SDZWA-v1`#	Classification	San Diego Zoo Wildlife Alliance	Americas	51 species • Trained on 556k images
`GIF-JAP-v0-2`#	Classification	Gifu University	Asia	13 classes • prototype on limited, imbalanced data • very biased to deer
`HWI-ADS-v1`#	Classification	Addax Data Science	Americas	AI Puaʻa v1.0 • 15 classes • USDA Forest Service & TNC
`VIC-ADS-v1`#	Classification	Addax Data Science for Parks Victoria	Oceania	212 classes • Trained on 5M images
`SBUSA-ADS-v1`#	Classification	Addax Data Science	Americas	68 classes • 3M training images
`AHDRIFT-v1`#	Classification	The Ohio State University, Columbus Zoo and Aquarium, Addax Data Science	Americas	Built for top-down close-up cameras
`IND-ADS-v1`#	Classification	Addax Data Science	Asia	Fine-tuned SpeciesNet • 40 classes • Trained on 1.1M images
`ANT-ADS-v1`#	Classification	Addax Data Science	Oceania	Fine-tuned SpeciesNet • 140 classes • Trained on 890k images
`EUR-DF-v1-5`#	Classification	The DeepFaune initiative	Europe	40 classes • September 2026 • added mongoose, arctic fox
`EUR-DF-v1-4`#	Classification	The DeepFaune initiative	Europe	38 classes • October 2025 • added golden jackal, raccoon dog, porcupine, muskrat
`EUR-DF-v1-2`#	Classification	The DeepFaune initiative	Europe	30 classes • October 2024 • added beaver, fallow deer, otter, raccoon
`EUR-DF-v1-1`#	Classification	The DeepFaune initiative	Europe	26 classes • February 2024
`PAN-SDZWA-v1`#	Classification	San Diego Zoo Wildlife Alliance	Americas	Andean highlands • 53 classes
`QLD-WOB-v1`#	Classification	Prakash Palanivelu Rajmohan and Renuka Sharma	Oceania	Wet Tropics rainforest • 15 classes
`AWC135-AWC-v1`#	Classification	Australian Wildlife Conservancy	Oceania	Continent-wide • 135 classes
`NZS-WEK-v3-03`#	Classification	wekaResearch	Oceania	Native + introduced • 81 classes
`NEO-MNCN-v1-0`#	Classification	Andrea Zampetti, National Museum of Natural Sciences (MNCN-CSIC), Madrid, Spain	Americas	Central + South America • 84 classes
`AFR-DFV-v1`#	Classification	Hugo Magaldi - One Forest Vision initiative	Africa	Congo Basin rainforest • 34 classes
`AFR-DFV-v2`#	Classification	Hugo Magaldi - One Forest Vision initiative	Africa	Congo Basin rainforest • 61 classes • DINOv3
`CAM-AI4G-v1`#	Classification	AI For Good Lab, Microsoft	Americas	Amazon rainforest • 36 genera • PytorchWildlife
`SAC-DMO-v1`#	Classification	Dan Morris	Americas	Top-down close-up cameras • 29 classes
`SOCAL-IRC-v3-6`#	Classification	Irvine Ranch Conservancy	Americas	Orange County • 18 classes • Irvine Ranch Conservancy
`NZI-ADS-v2`#	Classification	Addax Data Science	Oceania	Best overall • 24 classes • DOC New Zealand
`MD5A-0-0`#	Detection	Dan Morris	—	Industry standard • Battle tested
`MD5B-0-0`#	Detection	Dan Morris	—	Alternative to 5a • Dataset-dependent performance
`MD1000-REDWOOD-0-0`#	Detection	Dan Morris	—	Promising new standard • Not battle tested yet
`MD1000-SPRUCE-0-0`#	Detection	Dan Morris	—	Low accuracy (-14%) • 13x faster than 5a
`MD1000-CEDAR-0-0`#	Detection	Dan Morris	—	High accuracy (-1%) • 2x faster than 5a
`MD1000-LARCH-0-0`#	Detection	Dan Morris	—	Good accuracy (-3%) • 2.4x faster than 5a
`MD1000-SORREL-0-0`#	Detection	Dan Morris	—	Moderate accuracy (-3%) • 7x faster than 5a
`DINOV2-VITS14`#	Embedding	Meta AI (FAIR)	—	Fastest • 384-dim • Recommended for most projects
`DINOV2-VITB14`#	Embedding	Meta AI (FAIR)	—	Balanced • 768-dim • ~2.5x slower, finer detail
`DINOV2-VITL14`#	Embedding	Meta AI (FAIR)	—	Best quality • 1024-dim • ~7x slower, GPU recommended

Which one should you pick​

Which one should you pick