Ultra-wide Angle Lenses
Last updated: Man-O-War, 1 Nov 2010
Here is some analysis of the "MTF (resolution)" data for a number of ultra-wide angle (UWA) lenses from the test reports found on photozone.de.  It includes the following lenses:
Ref Test Data Notes
Lens Body
Canon 24-105mm f4L 350D (8MP) Not UWA, but I have one and included it for comparison
Canon 24-105mm f4L 50D (15MP)
Canon 24-105mm f4L 5DII (21MP)
Canon 10-22mm f/3.5-4.5 350D (8MP)
Canon 10-22mm f/3.5-4.5 50D (15MP) Probably close to my use on the T2i, an 18MP sensor
Sigma 8-16mm F4.5-5.6 DC HSM 50D (15MP)
Sigma 10-20mm F4-5.6 EX DC HSM 350D (8MP)
Sigma EX 10-20mm f/3.5 EX DC HSM D300 (10MP)
Tokina 11-16mm f/2.8 350D (8MP)

Key to the data in the charts below:
E.g. C10-22/15MP/10/C = Canon 10-22mm lens tested on a 50D (15MP)
body at 10mm focal length, center of the image
   [Manufacturer] [Lens-focal-length] / [m]MP / [nn] / [a]
  • [Manufacturer] = C for Canon, S for Sigma, or T for Tokina
  • [Lens-focal-length] is the lens's focal length range, e.g. "10-22"
  • [m] is the MP of the camera used in the test, 8MP was a Canon 350D, 10MP was a Nikon D300, and 15MP was a Canon 50D
  • [nn] is the focal length setting for that set of data points
  • [a] = C for Center, B for Border, or E for Extreme Border
LW/PH = line widths per picture height "which can be taken as a quantity for sharpness"
Aperture = the range of aperture settings for that set of data points
The subjective ratings (ex, very good, good, and fair) comes from photozone.de

Preliminaries: Effect of sensor size in these test results
One of the interesting factors in these tests is the effect of the resolution of the camera on the results.  Ideally I think you would want to test using a camera body whose sensor is not the limiting factor.  And do all tests using the same camera body.  That's not practical because the camera sensors are continually improving.  The folks at photozone specify in their reports which body they tested with and much to their credit, repeated the test for some of these lenses (the Canon 10-22 and Canon 24-105) on several generations of camera bodies (the Canon 350D, 50D, 5DII - with 8, 15, and 21 MP sensors respectively).
This first pair of charts compare an 8MP sensor (350D - the solid lines) with a 15MP sensor (50D - the dotted lines).  The chart on the left is at the wide end (10mm), the one on the right is at the long end (22mm).  All the data is using the Canon 10-22:
Moving up in sensor pixel count makes quite a difference.
Here is a similar comparison, this time with the Canon 24-105 - and now including a 21MP sensor (5DII).  Wide end (24mm) is on the left, long end (105mm) on the right.
Again, quite an increase with the higher MP sensor.
Here is another way of looking at the same data.  Each thin line is a separate combination of zoom-setting/f-stop setting across the 3 camera bodies.  Center and Border of the image is also diferentiated.  For example, one line might be the data for [ 24mm, f/4, Center of the image ] across the 3 bodies.  The thick red and green lines are the averages for the 24mm and 105mm data sets.
  The results (y-axis) are all relative to the 350D data point.  So, the 50D average resolution is 1.24x or 24% higher than the 350D.  And the 5DII average resolution is 1.67x or 67% higher than the 350D.
BTW, the 2 green lines that
rise a little above the rest are
'105mm f/4 Center' and
'105mm f/5.6 Center'.

The number of rows of pixels for each of these sensors are shown to the right.
The 1.63x for the 5DII vs 350D looks to be in line with the 1.67x average increase shown above.  The 1.24x increase of the 50D seems a little low though, considering the 1.38x increase in the number of rows of pixels.  I assume that's because the 50D sensor is much denser (5.4 MP/cm² compared with 2.4 for the 350D) reducing the 50D individual sensor pixel's light-gathering ability.
BodySensor #
of rows of pixels
Relative to 350DPixel density
350D23052305/2305 = 1.00x2.4
50D31683168/2305 = 1.38x5.4
5DII37443744/2305 = 1.63x2.4

Assumming all else being equal (same test procedure, same lens, same post processing of the images, etc), the results are very much dependent on the body and sensor size.  Thus compare results between tests that used the same body.
OK, so on to comparing UWA lenses
So for an apples-to-apples comparison (w.r.t. sensor size), here are the results using the available test data for the Canon 50D (a 15MP sensor):
The Canon 10-22 and Sigma 8-16 look pretty close, with the Sigma maybe slightly edging out the Canon.  The Canon is faster though, starting at f/3.5 vs the Sigma's f/4.5.  The Sigma is 2mm wider (it doesn't sound like much but it is 20% compared to the 10mm of the Canon).  The Canon is 6mm longer, pretty well filling the gap to my 24-105mm lens, FWIW.  The Sigma has a 121° angle of view while the Canon is at 107°.  Hmm.
Here are all the results using the available test data for the Canon 350D (an 8MP sensor):
The Tokina looks pretty good - not only faster (at f/2.8) but also apparently sharper.
My decision w.r.t. buying an UWA lens
After mulling it over a while, I bought the Canon 10-22mm f/3.5-4.5 - mainly I think because of what I hope is better piece-by-piece-off-the-production-line quality control and (based on this article) it appearing that the initial design/production bugs having been well worked out by now.  Since I retired, I live outside the US and returning a "bad copy" of a lens would be a nuisance (as well as expensive).  Here are some photos taken with the lens.  And a couple "big sky" photos.
More reading:
  Do Sensors “Outresolve” Lenses?