The reason for the GTX480 being faster is that it has 15 SM compared to 14 from the Tesla 2050. Also the GTX 480 runs at a higher clock speed (700 compared to 575). Put together this is 575/700*14/15 = 76.7% which comes pretty close to the 75%.
This is where the computer works. Since the original pinhole pattern is known (you created it) you can feed that pattern into the computer and it can use that pattern to "slide" all the overlapping images so they exactly fit on each other. This makes a single, bright, sharp image.
The computer not only deconvolutes the pinhole pattern(which only provides an image with the same resolution as the pinhole) but it uses the entire diffraction in the detector and phases it to obtain an image, much like a lens would do. This achieves a resolution that is simply limited by the numerical aperture of the detector (which can be much smaller than the pinhole size).
2500 times better signal/noise, or (I think) 50 times better resolution, or 2500 times shorter exposures...
The improvement in resolution will be much smaller than 50 times as the intensity drops very fast with increasing resolution. It's not simply the sqrt of the SNR.
By post processing they mean the phasing algorithm. The imaging method is divided in two parts.
First they recover a low resolution image of the object that was imaged by looking directly at the hologram and deconvoluting with the known pinholes (in this case a Uniformly redundant array (URA), which assures that the deconvolution is well behaved). This step cannot achieve a resolution higher than the size of the pinholes in the URA.
In a second step the entire image is phased, meaning that an algorithm is aplied to it that tries to mimick a lense. This increases the resolution obtained to the maximum possible, that is to the limit of the numerical aperture.
Slashdot Mirror
The reason for the GTX480 being faster is that it has 15 SM compared to 14 from the Tesla 2050. Also the GTX 480 runs at a higher clock speed (700 compared to 575). Put together this is 575/700*14/15 = 76.7% which comes pretty close to the 75%.
By far Apple ain't biggest in IT, they are way smaller compared to some other companies. Say, HP, Dell, Microsoft, Nokia.
Apple is the largest company, by market capitalization, from the ones you mentioned.
http://www.pnas.org/content/early/2009/01/12/0812068106.abstract
This is where the computer works. Since the original pinhole pattern is known (you created it) you can feed that pattern into the computer and it can use that pattern to "slide" all the overlapping images so they exactly fit on each other. This makes a single, bright, sharp image.
The computer not only deconvolutes the pinhole pattern(which only provides an image with the same resolution as the pinhole) but it uses the entire diffraction in the detector and phases it to obtain an image, much like a lens would do. This achieves a resolution that is simply limited by the numerical aperture of the detector (which can be much smaller than the pinhole size).
2500 times better signal/noise, or (I think) 50 times better resolution, or 2500 times shorter exposures...
The improvement in resolution will be much smaller than 50 times as the intensity drops very fast with increasing resolution. It's not simply the sqrt of the SNR.
By post processing they mean the phasing algorithm. The imaging method is divided in two parts. First they recover a low resolution image of the object that was imaged by looking directly at the hologram and deconvoluting with the known pinholes (in this case a Uniformly redundant array (URA), which assures that the deconvolution is well behaved). This step cannot achieve a resolution higher than the size of the pinholes in the URA. In a second step the entire image is phased, meaning that an algorithm is aplied to it that tries to mimick a lense. This increases the resolution obtained to the maximum possible, that is to the limit of the numerical aperture.