Dear Wildbook Community,
We are now pleased to announce that the new MiewID v4 AI model for individual animal re-ID is ready for Wildbook deployment!
MiewID - developed by our Wild Me Lab here at CXL - is the state of the art AI model for visual mark-recapture, and it directly benefits from the collaboration inside Wildbook and data pooling across the many diverse species supported by Wildbook (currently 284). We are always honored to be trusted with your data, and we are able to return that trust with ever-advancing and improving AI models to benefit your research and conservation efforts.
MiewID v4 represents a big leap in training data, species coverage, and performance. Compared to the v3 version, MiewID v4 was trained on 40% more data and species from public and private mark-recapture catalogs for approximately 90 species and 110 feature classes (heads, flanks, fins, flukes, faces, etc.).
Averaged across all species, MiewID achieves:
Avg top-1 ID prediction: 78% (+2% improvement from v3)
Avg top-5 ID prediction: 87% (+3% improvement from v3)
Avg top-10 ID prediction: 89% (+4% improvement from v3)
Almost all species saw consistent or improved performance in v4 training, while some species saw dramatic improvements, such as:
- Iberian lynx (+12%)
- Eurasian lynx (+12%)
Some of the new species included in the v4 training data are:
- African elephants
- Asian elephants
- Chital deer
- Sika deer
- Cattle
- Yellow-bellied toads
- Tiger sharks
- Spotted lantern flies
- Southern right whales
- Bowhead whales
- Degus
- Domestic cats
- Domestic dogs
- Capuchin monkeys
- Fire salamanders
- Sandtiger sharks
- Leafy seadragons
- Weedy seadragons
- Southern giraffe
- Northern giraffe
- And more.
We are excited to explore how v4 generalizes to improve existing species ID prediction in Wildbook and offers “zero shot” matching for new species, such as for our expanded coverage for 25 species for small, understudied wild cats in Whiskerbook.org.
We will provide species-specific breakdowns for v4 in an upcoming preprint. You can see some performance on past species in the MiewID v3 (64 species) results:
MiewID’s original architecture (v2) is documented in:
Otarashvili, L., Subramanian, T., Holmberg, J., Levenson, J. J., & Stewart, C. V. (2024). Multispecies animal re-ID using a large community-curated dataset. arXiv. https://arxiv.org/abs/2412.05602
Over the next two weeks, our team will be replacing MiewID v3 with v4 and bringing you the state-of-the-art for photo ID of wildlife study populations. Please look for a banner inside Wildbook after login to announce the change for your Wildbook(s).
If you have any concerns about migrating your species from MiewID v3 to v4 in Wildook, please let us know.
MiewID is faster and more scalable than the older HotSpotter algorithm, allowing us to run more AI for more species every day, which is important as Wildbook experiences growth in its userbase (+25% in 2025), datasets (+22% in 2025), and species coverage (+40% in 2025). And we’re excited to continuously improve the accuracy with which AI can distinguish individual animals.
Your collaboration in Wildbook enables great things for a global community, and we look forward to providing Wildbook as an ever-improving resource for your benefit as part of our long-term, nonprofit mission.
Dare Mighty Things,
The Wild Me Team at Conservation X Labs