New AI model released

171 views
Skip to first unread message

Wildlife Insights

unread,
Apr 18, 2022, 2:50:23 PM4/18/22
to Wildlife Insights

A new AI model is now available in Wildlife Insights! We’re sharing updates on the model and tips on how to use model results and platform features to save time reviewing images.  

Here’s what’s new:


A larger training dataset

Our training data now encompasses 993 species (+2 non-animal classes and 167 higher taxonomic classes) across 15 million images. That’s up from 886 species and 12 million images. Check out the species table to see how the model performs for each species in the training dataset. 

 

Improved blank image recognition

This new model weights results from a species classification model trained on data from Wildlife Insights and a modified version of the MegaDetector object detector. This ensemble approach increases the number of images the model can confidently classify. Compared to the last model, this new model identifies more blank images and is less likely to incorrectly identify an object as a blank, minimizing the possibility of removing valuable images of animals.

 

More animal predictions

Across the board, this model is more confident recognizing if there’s an animal in the image or not. This means that there are fewer images returned with “No CV Result” -- the result shown when the model isn’t confident at any taxonomic level. 

 

How to use AI to accelerate image review

The Wildlife Insights AI model is trained to maximize efficiency by minimizing the time you spend looking at blank images. 

We’ve calibrated the model to classify an image as blank only when it’s confident, but AI models (like humans) are prone to mistakes. The Wildlife Insights model may occasionally mark an image as blank when it actually contains an object or animal of interest. Luckily, using the tools in Wildlife Insights can help you review blank images in bulk and quickly process thousands of images in just a few minutes. 

Here’s how:

  1. Go to the Identify tab in your project.

  2. Apply the “Blank” status filter as a first step to display all images marked as blank by the AI model.

  3. Increase the thumbnail size to medium to see a larger preview of each image.

  4. Select the option to view 200 images per page.

  5. Scroll through the page to review each thumbnail. 

  6. If all images are blank, you can use shortcut keys to select all images on the page (Ctrl+A or Cmd+A) and mark the images as blank. 

  7. If any image appears to have an object of interest, select that image, edit the ID, save the ID to send the image to the Catalogued tab, and return to the thumbnail view to review blanks.

 

We’re eager to continue improving the model by adding to the breadth of the training dataset. You can help out by continuing to upload your images and verify the identifications in Wildlife Insights. 

If you have any questions or feedback for us, please visit our About our AI page to learn more or post a question in this Community Forum. Thanks for your support!


Reply all
Reply to author
Forward
0 new messages