<div dir="ltr">Here's another cool ECCV paper. Mathieu in particular may find it interesting since it involves fMRI<div> <div><a href="http://web.stanford.edu/~nsid/publications/barbu2014fmriAction.pdf">http://web.stanford.edu/~nsid/publications/barbu2014fmriAction.pdf</a><div>
<br></div><div>Summary: They perform action recognition two ways. The first way uses standard vision techniques to train a classifier for action recognition from videos. The second has a human observer watch the same videos while undergoing fMRI. The brain scans are then used for training a classifier for the videos. They compare these results to each other and it turns out that using the brain scans yields much better results.</div>
<div><br></div><div>An interesting consequence is that this implies computers are much better at reading our minds than they are at recognizing actions in a video...</div></div></div></div>