Ultra-fast categorization of image containing animals in vivo and in computo

Animals are able to categorize images from broad categories in a very efficient and rapid fashion. Humans, for instance, can detect the presence of an animal in an image in as little as 120 ms. In the last decade, the field of artificial intelligence has experienced one remarkable breakthrough. In a relatively short period of time, neuroscientifically-inspired deep-learning algorithms designed to perform a visual recognition task literally bloomed.  Artificial networks now achieve human-like performance levels, but are usually trained on less ecological tasks, for instance the 1000 categories of the ImageNet challenge. Here, we retrained the VGG Convolutional Neural Network adapted to ImageNet on two ecological tasks : detecting animals or artefacts in the image.

This zotero group aims at gathering litterature on the subject, both from biology and machine learning.