The Wildlife Insights platform uses AI technology to drastically reduce how long it takes to convert camera trap data into usable biodiversity insights. Anyone who collects camera trap data can use Wildlife Insights to upload data to Google Cloud and access AI models trained to automatically classify camera trap images. The Wildlife Insights AI model is trained using Google’s open source TensorFlow library.

Here's how AI can help:

The Wildlife Insights AI model is trained with the largest curated dataset of labeled camera trap images to date, with over 35 million images across 1,295 species. You can help -- adding your data to Wildlife Insights can help expand our AI capabilities to new species, existing rare species and improve our accuracy on the species already in the dataset.

Check this table to see if a species is in our dataset and how well the model predicts it.

# of images: this is the number of images that went into the training data set for the latest version of the model.

Precision refers to how many of the model's predictions were correct. A high precision for a given species means that the model tends to be right when it predicts that species.

Recall refers to the number of images that the model correctly identifies for a species out of all the images of that species. Note, that if the model isn’t confident about the exact species, it may produce predictions at higher levels of the taxonomy (e.g. Mammalia, Animal). 

Unknown rate: Sometimes we know that our AI models are less confident about their own predictions, and instead of making risky, potentially wrong predictions, we mark those as "unknowns" (or "No CV Result"), so you know they need to be manually reviewed. In the table above we report the unknown rate for each species, which is the proportion of unknowns you can expect to receive from our model.

You can find examples of precision, recall, and unknown rate applied to Blue duiker, In the FAQ Section below, under Accuracy/How is the confidence of the model evaluated?

If the number of images for a particular species is less than 1000, this means our AI model is not able to make a correct prediction with confidence above a fixed threshold (the threshold is tuned manually). If you do not see your species of interest listed it means that we currently have no examples of that species in our dataset. This is all the more reason to contribute, so we can continue to grow the number of supported species in Wildlife Insights.

Learn more about assessing classification accuracy for AI models in general.