News for hackers. Stuff that matters.
Re:Building a deeper understanding of images by hyperion2010 on Friday September 05, 2014 @01:57PM (#8275259)
I wonder whether some of the intermediate layers in these models might correspond to something like "living room" or other locations that provide additional information about the objects that might be in the scene. For example, I suspect it was much easier for me to identify the preamp and the wii in one of the pictures because I knew it was a living room/den instead of an office or study.
Re:Building a deeper understanding of images by pdenya on Friday September 05, 2014 @06:05PM (#8276561)
In this case, no they didn't have pre-labeled location/setting available. You can see one of the datasets they used here: http://image-net.org/challenges/LSVRC/2013/#data.

Generally speaking, Neural Networks are black boxes. The layers interact with each other but not in a defined categorical manner like that. Layer size/depth are parameters you provide when setting up that have tradeoffs in result accuracy, space, time spent, etc like jpeg quality.