Validation is performed by taking a new image (not used in training the neural network), collected during a field session, and matching it to a catalogue of images collected from previous field sessions. ID's of the catalogued images were assigned by field personnel familiar with the population of dolphins from that particular bay. These personnel look at the images returned by finFindR and verify if the top image from finFindR is a correct match or if a correct match is in the top 10 images returned etc.

Barataria Bay September 2017:

• 88.50% top 1 accuracy (354/400)
  
• 93.25% top 10 accuracy (373/400)
  
• 97.25% top 50 accuracy (389/400)
  

Barataria Bay May 2017:

• 87.50% top 1 accuracy (238/272)
  
• 94.12% top 10 accuracy (256/272)
  
• 96.69% top 50 accuracy (263/272)