This is our May 2015 top neural network for large-scale object recognition. It has been trained to recognize most typical home indoor/outdoor objects in our daily life. It was trained with more that 10 M images on a private dataset. It can serve as good pair of eyes for your machines, robots, drones and all your wonderful creations!
See it in action in this video #1, and also this other video #2.
This application is for tinkerers, hobbiest, researchers, evaluation purpose, non-commercial use only.
It has been tested on OS X 10.10.3 and Linux. It can run at > 17 fps on a MacBook Pro (Retina, 15-inch, Late 2013) on CPU only.
Install Torch7: http://torch.ch/
Please download files: model.net, categories.txt and stat.t7 from https://www.dropbox.com/sh/qw2o1nwin5f1r1n/AADYWtqc18G035ZhuOwr4u5Ea?dl=0
Linux camera install: cd lib/ then make; make install. Note that Makefile wants Torch7 installed in /usr/local/bin, otherwise please change accordingly!
To run with a webcam and display on local machine: qlua run.lua
Zoom window by 2 (or any number): qlua run.lua -z 2
Feel free to modify and use for all you non-commercial projects. Interested parties can license this and other Teradeep technologies by contacting us at [email protected]
Have fun! Life is short, we need to produce while we can!
Aysegul Dundar, Jonghoon Jin, Alfredo Canziani, Eugenio Culurciello, Berin Martini all contributed to this work and demonstration. Thank you all!