Focusing the Sigma 8mm F3.5 EX DG Circular Fisheye

I have been using this lens with a Canon EOS 400D. Reading this article, I decided that it made sense to focus the lens at infinity, so the lens was set to manual focus at infinity and taped down. This way everything at a far distance is in focus, and anything larger than 2 mm will be recognizable as long as the lens can resolve it. I was fairly happy with the results, but felt that I had to stop down more than should have been necessary.

The 400D was eventually replaced by a 40D. The 40D has live view, and I decided to have a look at the focusing of the Sigma. I autofocused it using AI Servo on something far away (the barrel stopped almost at the infinity mark), switched the lens to manual focus and checked the live view at 10x magnification. It was somewhat sharp, but not as good as I had expected. Then I adjusted the focusing manually until it was as sharp as possible. I looked at the distance marking, and was flabbergasted. It was set at just this side of 1 m! The test was repeated several times - autofocus would set the lens at infinity, manual refocusing would set it at just under 1 m. (Tests with auto focus set for one shot would place focus all over the place - the AI servo sort of 'pulled in' the focus to a common value.)

I tested several Canon lenses too, none of them had this problem. It is possible to get more exact focus by using live view, but none of them were consistently far off like the Sigma was. See the article at Nature Photographers for an alternative opinion on the use of manual focus (preferrably with live view) vs. auto focusing.

I posted the findings at panoguide, and in the discussion that followed, several posters gave me valuable advice, corrections and links, which set my thinking straight so that I could acheive optimum results from the Sigma. Thanks to all that provided feedback!

Diffraction and Circle of Confusion

First some words on diffraction, which is very well explained on this page. The 40D is diffraction limited at about f/11, at which point the diameter of the Airy disk is close to two pixels. (In the graphics about half way down the page, the camera type Nikon D2X has a pixel size that is very close to the Canon 40D.) Using an aperture smaller than f/11, will result in greater depth of field, but less absolute sharpness. The Sigma should be used in the range f/4 to f/8 for optimum sharpness.

Circle of Confusion (CoC) is the name of the blurred disk that an out of focus point represents on the sensor. The maximum acceptable size of this disk determines the Depth of Field (DoF) that the image will have. There are any number of pages on the web describing this, a good starting point is the article at Wikipedia.

In panorama photography, every pixel counts. Far away objects are rendered so small on the sensor, that no blurring by diffraction or a too large CoC is acceptable, as this will reduce sharpness where it counts in most panoramas. By using f stops in the range f/4 to f/8 (or f/11 at the extreme), diffraction is kept in check. But how big can the CoC be?

The pixel size on the 40D is 5.7 µm. But a single pixel will only record one colour (red, green or blue (bayer pattern)). In order to capture all three colours in each pixel, the neighbour pixels must be analyzed to compute the values for the missing colours. To avoid aliasing effects, an anti-aliasing filter is placed in front of the sensor. The effect of this filter is to blur the image slightly, so that a single point in the image is recorded by more than one pixel, and thus ensuring that all three colours for the point can be reconstructed from neighboring pixels.

Because of this filter, a CoC diameter can be larger than one pixel, but probably not larger than two pixels before sharpness is affected. A CoC of 10 µm (0.01 mm) is 1.75 times larger than a pixel, and can be used in a depth of field calculator. (The CoC size used for the markings on the Sigma seems to be 0.02 mm.)


Two methods can be used to find the correct focus position to tape down the foucus barrel on the Sigma at.

In the first method, use the live view on the camera, or preferably remotely controlled on a computer (or by taking a series of pictures, if you do not have live view). Start by focusing at infinity, and slowly focus closer and closer. At a certain point, maximum sharpness is achieved, and there is little or no apparent change in sharpness for a while until you get to a point where sharpness starts to decrease again. This last point is the hyperfocal setting where you want to tape the focusing barrel at. The 'no change' section is where the CoC is smaller than 10 µm. When doing this on the camera, it is natural to use the maximum aperture - on a PC (or with a sequence of pictures) f/4 is a good choice.

In the second method, the depth of field calculator is used to calculate the hyperfocal distance. For f/4 and CoC 0.01 mm, this evaluates to 1.61 m. Place the camera at this distance from something distinct to focus on (measured from the sensor, see marks on top of the camera body). This time the correct setting is the center of the 'no change' section on the focus barrel.

By using f/4 in these tests, you ensure good sharpness at infinity, and even better close focusing ability when stopping down more. Since auto-focus cannot be trusted, and the focus barrel is very loose in manual focus, you really want to tape it down to avoid moving the barrel accidentally, or even forgetting to set it correctly.

The bottom line is that there are enough people that have confirmed this 'flaw' in the Sigma 8/3.5, that you owe it to yourself to do these tests on your copy of the lens if you have not done so already.

100% crops from the center of images taken at f/4.0. In the sharp image, the focusing barrel has been set in the center of the 'no change' region for maximum sharpness.

Return to the panorama page.