Computers

• An important computer vision resource is a set of boxes we call "Computational Sensors". The idea of a computational sensor is to put a lot of computation close to the camera to perform fast preliminary processing, and hook it into a fast network so that additional processing power (e.g. a cluster or high-end multiprocessor) can be brought to bear if needed. The current basic model is a dual processor Linux box with lots of main memory, a big chunk of local RAID disk storage, (intended to store video data short term), and gigabit ethernet connections. There are currently three generations of these machines in the department. All utilize dual Intel Xeon processors (the Xeon is basically a Pentium III with additional level 2 cache and some other extras), 1GB of main memory, and 280GB of RAID disk storage. Generation 1 boxes have shark names (gulper, wobbegong etc.) and 1.5 GHz processors. There are 4 of these in the vision lab, set up as workstations, and intended to have user-selected cameras attached as needed. Generation 2 boxes have shellfish names (Lobster, Mussel, etc.) and 2.0 GHz processors. There are 4 of these in a bank in the vision lab, with ceiling-mounted cameras connected to them. These are intended to be stable connections. The bank machines are typically accesed remotely, though there is a terminal switch connected to the bank. Shrimp is set up as a workstation. Generation 3 boxes have whale names (humpback, bowhead etc.) and 3.0 GHz processors. There are 3 of these in the software lab, each connected to a pair of ceiling-mounted cameras. 4 others will eventually be placed in a bank in the vision lab, along with the generation 2 machines.
• There are two old Solaris multiprocessors, important mainly because they are currently the only way we can control the robot arms and the hand. Both are shut off, but might work if turned on. Brain is a Sun multiprocessor with 4 167 MHz processors, and 2 gigabytes of core. It has one or two KTV digitizers attached. It was originally intended for running experiments requiring large amounts of memory, but the aorta and the alpha cluster now have even more. It is getting a bit long in the tooth now but might still be useful for applications needing two digitizers, and has enough memory to make it useful for digitizing video sequences, even if it can't quite handle frame rates.
• Aorta is a newer Sun multiprocessor with 8 336 MHz processors and 3 gigabytes of core. It runs the robot peripherals in the lab, including the two puma arms, the Utah hand, the Maxvideos, and miscellaneous other equipment. It also has two KTV digitizers attached. The idea is to provide a central resource that can coordinate a number of peripherals for experiments in hand-eye coordination etc. It could also be used for experiments requiring large amounts of computation or memory, but check to make sure it is not running some real-time experiment before loading it with some time-hog program.
• There are also the department multi-processors and clusters, which are available for computational purposes, though getting dedicated use would take some scheduling. These include the IBM Regatta machine (32 power-4 processors with integrated shared memory), two Sun multiprocessors, (a 16 and a 32 processor machine), and the department linux cluster (64 dual-pentium4 pizza boxes). Quite a set of resources if you can think of something to do with them.
• Then there are two SGI Onyx infinite reality engines, graphite and silicon, which can produce texture-mapped graphics for complicated scenes at frame rates. These are used to drive VR equipment in the virtual reality lab, but are also available for other applications. The older box, graphite, is particularly underutilized. If you really want to do graphics, these boxes are for you. Of course you will have to figure out how to use them. They may not be around much longer, as much of the VR lab equipment is leaving with Dana.

Back to vision course main page