canon 8-15 vs nikon 10.5 ,tokina

已有 10238 次阅读2012-7-20 08:07 AM |系统分类:拍摄散记| tokina, canon

Review of the Canon EF 8-15 mm f4 Fisheye Zoom A new tool for the Canonist panorama photographers. Report from tests with a full frame Canon DSLR New (Mar 2012) You may now read a supplement to this page about "What are the competitors?"


Table of content:


Previous other reviews on the web

This lens is designated by Canon as part of the 'L" series (Pro). It has been made available at the very end of July 2011. Many reviews were issued a few days or a couple of weeks after that. AFAIK the reviewers have been using the new lens for "normal" photography.

I have particularly appreciated and recommend the reader of this article to read the reviews by , and by

A nice overview was posted by Canon USA in November 2010...

My intention is thus to supplement such reports and to give my point of view as a panorama photographer, and not to dupplicate them. I shall focus the report on panorama shooting specific observations when possible. But in addition and since many stitched panorama source images are currently and conveniently shot with a FE (fisheye) or an Ultra-Wide-Angle (UWA) lens, I shall attempt to confront the newest Canon zoom with other lenses that are frequently used by the panorama photographic community.


Canon has designed a product that can be used on a Canon DSLR of any sensor size (APS-C, APS-H and full frame. Read second link above). I have been using the new lens (almost) only on the FF EOS 5D Mk 2 for the moment. Consequently and at this early stage, this article belongs only to this context. I suggest the user of a smaller ("cropped") sensor cameras to read it anyway and to come back later on. I plan to complement ASAP the report with a Part 2 ASAP....

No-Parallax Point

Despite (or because?) nearly 2-3 millimeters of translation of the front group of optical elements when zooming in or zooming out (it is most retracted at around 11 mm of focal length), the NPP (a.k.a. Least Parallax Point) doesn't move when zooming the lens! In practice, the NPP moves so little along the axis that it can be considered as a stationary point. This holds true unless you would attempt to take a "2-shots" or a "3-shots" panorama and with objects very close to the front of the lens. This means that both the Tokina 10-17 mm and the new Canon lens are fisheye zoom lenses that share the nice feature of a "nearly stationary NPP" along the whole focal length range.

Of course and as for all the other fisheye lenses (as well as retro-focus UWA), the NPP location moves along the axis as a function of the selected or required incident angle. In the practice of panorama stitching ,that generally depends on the number of shots to cover 360° around. The shift movement of the NPP extends along a 15 mm distance on the axis of the Canon 8-15 mm lens.

Assuming four images around, the NPP location is illustrated on the picture below by the green dot:

Three other alternate locations are also indicated (in brackets). At 15 mm setting the front element is protruding the most from the lens barrel and it's ~3 mm forward of the position it gets when zooming is set at 11 mm (i.e. the most retracted position). This movement of the front optical group slightly changes the NPP location for two shots and also (but in a lesser extent) for three shots. More than 3 shots around are needed when the focal length is set at 15 mm...


Focus distance, autofocus and hyperfocal point

I have found this subject to be the main hurdle when using the lens, especially if one wants to shoot with it wide open (at f/4). Blurring happens more often than not if care isn't taken to properly adjust the focus distance setting. Much less than half a millimeter of shift on the focus distance scale shall put the picture out-of-focus. 0.5 mm corresponds to a rotation of the manual stting ring of about 1 to 1.25 mm only!

The image is homogeneously sharpest when aperture is set at about f/5.6 or f/8. This aperture setting is IMO possibly the best compromise (depth of field, sharpness from center to corner, etc.) for panorama when lighting of the scene and subject motion velocity allows to shut the diaphragm down to such values.

By trials and errors, I have found the hyperfocal point to be best set when the "right edge" of the vertical "leg" of the digit "1" (1 meter) is aligned with the vertical distance indicator on the barrel of the lens. At f/8, everything in the object space located from about 70 cm to infinity from camera is crisp focused on the sensor plane.


1) Quote from the User's instructions Manual: << The infinity position at normal temperature is the point at which the vertical line of the L mark is aligned with the distance index on the distance scale. >> I honestly cannot confirm this affirmation. The infinity position is definitely NOT at this position at "normal temperature", at least on the lens unit that I have in hands. The infinity position is most probably just "slightly before" (i.e. out on the left of) that vertical line of the L mark. This can be easily confirmed by carefully aiming the lens at "infinity" in real object space and in Auto-Focus mode (shutter switch half depressed) and subsequently noting the corresponding distance scale indication. If one puts the mark as Canon suggests, the whole photograph is blurred, including infinty... by out-of-focusing!

2) This observation is important because the distance between the "1" (one meter) and the infinity mark is ONLY about 1 mm on the scale. There isn't a way to manually set the exact distance of focus to a nearby object beyond one meter! That is nevertheless the most frequent shooting situation in panorama photography, isn't it? At this stage, let us quote : << To all intends and purposes a fisheye lens can make do without an autofocus. Working on full frame at 8 mm, even wide open, you can enjoy sharp images in the range from 0.9 of a meter to infinity when you set the distance scale at 1.8 metres. On stopping down the situation becomes even better. >>Hmm... setting the distance scale at 1.8 metres... is a nice suggestion. But, how on earth can you do that accurately? You may have a look on the scale on the picture of the lens right above and tell me how to do...

I dare to think that the suggestion is in practice equivalent to what was suggested above in this paragraph (i.e. "put the rigth edge of the #1 meter aligned with the distance indicator mark"). The 1.8 meter figure is actually the hyperfocal distance relevant to 15 mm of focal. This would also approximate a fit for the whole range of focal lengths but further refinement can be applied for the shorter end of the scale: for instance I thus set the mark aligned with the vertical leg of the "1" when I shoot with 12 mm of focal (my usual way).

3) Some may think otherwise, but I agree with the reviewers of the Canon new lens who say that fortunately, the AutoFocus is very fast and accurate on the new Canon lens (when lighting is sufficient, of course). Note that there are several Autofocus modes to choose from on the camera and that they are in many cases not all equally accurate. The 8-15 mm is a "L + USM" lens (i.e." pro" and "expensive") and the customer is expecting such a treat, isn't he? I then set the Autofocus by aiming at an object that is located ~1.8 meter from the camera and I switch the focus-setting mode back to manual.

4) To compensate for shifting of the infinity focus point that results from changes in temperature, on the Canon 8-15 mm lens there is an Infinity compensation mark (the horizontal leg of the bold "L" shaped mark). This means that one can willingly put the index (that is engraved on the lens barrel) up to about 5 mm beyond the Infinity point! Ironically, we may compare the meager 1.5 mm of distance between "1 meter" and infinity marks to the 6.5 mm distance between the same "1 meter" mark that applies for the very rare case where the infinity has shifted due to VERY wide temperature change... I would have preferred a more balanced proposition:-)

5) If you intend to put some tape stuck to fix the focus setting ring in place after adjustment of the hyperfocal you then must know that it is extremely important to also lock the Auto-focus switch to MANUAL and to put some tape on it. But unlike fisheye lenses that are not fit with piezoelectric actuation, the USM doesn't keep the mechanism really fixed. I have once been fooled by this system during a shot campaign at night and I got a... nicely blurred panorama. BTW I am not certain if masking tape application on the ring and on the switch after focus distance pre-setting (e.g. on the hyperfocal point), shall unsure this setting be kept really fixed if the camera is submitted to some moderately high white random noise vibration level....

REMARK: Canon (US) wrote in this article that is illustrated witth this image: <<For close-up photography at distances of 1.5 meters or less, the lens is designed so the equidistant subject circle (a circle described by an equal subject distance at all angles of view) appears flat in the image plane (see Figure). This allows the photographer to emphasize the center of the photo by focusing on the center and obtaining an attractive fuzzy effect at the edges when shooting a flat subject at close distances. Conversely, it is possible to capture a very clear image with the entire FOV in focus when shooting a subject that surrounds the camera (lens) at a uniform distance. When shooting subjects at distances over 1.5 meters, the lens is essentially pan focus, where nearly everything in the FOV is in focus.>> I have observed that this behavior (that in fact invalidate a frequent misconception) is true for most of the fisheye lenses and at the least for all of those that are dealt whith in this present article.

6) Tip: If, for any reasons, the Autofocus temporary solution -read item #3 hereabove- is not reliable, then viewing the scale of distance of focus (visible through the plastic transparent window) is essential to manually set the lens correctly and thus to get sharp-focused images. If for any reason, it is not visible (for instance if it's hidden inside a clamp ring), one has then to use a workaround to set the lens to the hyperfocal...

One simple way is 1) to push the setting ring manually so that it is mechanically end-stopped at {position >> ∞}and even farther so as to surely overshoot this position (the ring can be moved freely because the de-clutching effect of the USM), then 2) to reverse the rotation from this extreme end positon back to a final known position with respect to visible marks on the barrel. In other words, this final position is attained after rotation of a known fixed distance (i.e. 8.2 mm) from end-stopped state of the ring in order to set exactly the lens at the hyperfocal {H}.

As a matter of fact, this process is simulating a virtual "twin" distance-scale that replaces the obstructed original one. And this "new scale" is made available on another part of the lens barrel: in this effect, I have painted in white with a "white pencil" one groove on the rubber ring (rapid correction TippEx liquid may work equally). After having rotated the ring to overshoot beyond the infinity end stop action (position on the -invisible- scale at the Infinity compensation mark { >> ∞}), I move it back to the left edge of the letter "Z" of the word "ZOOM" and not more; (note that this has yet not changed th actual setting on the hidden scale itself). Then I reverse (again) the rotation and move the white groove to the right edge of the letter "M" of the same ZOOM word. The distance of focus indicator is now at this stage exactly aligned on the right edge of the "1" meter position on the scale and hyperfocal is set.

Remarks: the word "ZOOM"(see image above) is engraved on the front fixed part of the barrel. Any other adequate pair of points of reference on the body (e.g. two marks applied on the red thin ring) could be used as long as they are separated by ~8.2 mm. BTW there is a 5:2 ratio between the rotation of the ring and the synchronized distance scale and that de-facto magnification soothes in a way the difficulty to set accurately the hyperfocal. Voila!

Comparison with other lenses:

The difficulty to manually and correctly set the correct intended distance of focus is probably the cause of some early reporting (on forums) for observation of "softness" in some images that were shot with the new lens right out of the box....

Let's compare what measurement means that we can count on for setting the focus distance on some lenses that I may use for panorama:

Distance on the focus distance scale to the infinity mark...
.... from the "0.5 meter" mark
... from the "1 meter" mark
... from the "3 meters" mark

Sigma 8 mm f4

19 mm
9 mm
Samyang 8 mm f3.5
33 mm
16 mm
5 mm
Nikkor 10.5 mm f2.8
4 mm
Canon 8-15 mm f4
~ 2 mm (best estimate)
1 mm
Tokina 10-17 mm f3.5-4.5
~ 4 mm (best estimate)
Samyang 14 mm f2.8
62 mm
27 mm
8 mm

The Sigma 8 mm lens is (like the Canon zoom) also fitted with an Infinity compensation mark (the bold "L" shaped mark) for the same purpose as Canon explained it to be.

Neither the Nikkor 10.5 mm nor the Tokina 10-17 mm could actually be manually and accurately set at the hyperfocal just by using the scale:

  • The Tokina 10-17 mm lens (the model that I own) must be put at the infinity end (i.e. pushed to the mechanical stop) to get it perfectly set on hyperfocal (!) That's fast and reproducible but that's not what is expected.
  • The autofocus of the adapted-to-EOS Nikkor lens naturally doesn't work with EOS camera. The ring should be permanently fixed on hyperfocal by a piece of tape after careful calibration with a trial and error process. The distance scale was IMO the World worst-designed by Nikon... but it's now a tie with Canon. BTW compared with the usual "Canon Sandard", the scale is reversed on these two lenses (Tokina and Nikon).
  • Being lenses originally deprived of Autofocus ability, both the Samyang 8 mm and 14 mm had to be dismounted and mechanically adjusted to get the focus distance scale aligned about right. After this operation, the two lenses are the best and easiest amongst all to accurately and manually set the distance. I am certain that some testers have probably made wrong assessment of the lenses performances and IMO they gave a wrong jugement for instance here about the Samyang UWA lens: Ken contradicts the positive judgement that was reported by most -if not all- other test reviewers. He was probably fooled and did not reckognize the generic defective cause. Weak Quality Control at Samyang facilities is said to be the root cause of this annoyance.

By the way, there is often a misconception by many photographers about the

For close-up photography at distances of 1.5 meters or less, the lens is designed so the equidistant subject circle (a circle described by an equal subject distance at all angles of view) appears flat in the image plane (Figure 5). This allows the photographer to emphasize the center of the photo by focusing on the center and obtaining an attractive fuzzy effect at the edges when shooting a flat subject at close distances. Conversely, it is possible to capture a very clear image with the entire FOV in focus when shooting a subject that surrounds the camera (lens) at a uniform distance. When shooting subjects at distances over 1.5 meters, the lens is essentially pan focus, where nearly everything in the FOV is in focus.

Comparison of the coverage of images that were shot with the Canon zoom fisheye lens with images from other similar lenses 1) For a quick look and to get a subjective appreciation:

This (flash) object movie shows the differences that will exist between the images by different lenses.

Other reviewers have also posted many examples of animated images that illustrate the aspect of the different lens projections and different lens magnifications.

2) For inputting test images in your favorite stitching program(s): a panorama digital image Databank!

Many images from the new Canon lens... may be dowloaded in raw (.CR2) file format!

A table is indexing a large collection of files of full-scale images of the same scene (see thumbnail above). It allows the reader to select and download sets of test images that were shot with the EF 8-15 mm as well as four other fisheye lenses (and alternatively with a UWA lens). I must remind again the reader that all these images were shot with a full-frame Canon DSLR only.

Beware: some of the downloads are particularly bandwidth hungry!

Image radial mapping measurement and comparison with similar lenses

The projection that Canon has adopted for the new lens is.... normal for such a fisheye lens. The curves can be approximately stacked with those of similar fisheyes with Equi-Solid Angle type of projection. A chart demonstrates this feature:

Click on the thumbnail picture to enlarge the view. Or here to download a PDF document.

For the curious reader the experimental process that was used to get the graph is posted here. The radial compression when reaching the edge of the field is a bit less accentuated than most other fisheyes (except the Samyang, of course). It seems that Canon's own design specification was linear scaled: R= f x Theta (i.e. the equidistant projection) and that they finally got close enough to this target. Because of this feature, one may expect to get very small coefficients for image "distortion correction" after optimization in the stitching software.

Update (19 April 2012): Canon has filed in 2011 for various Patents applications at the US Patent and Trade Office that directly concern the 8-15 mm Lens. Two of these applications were published by US PTO early in 2012 that revealed graphs showing relative distortion of the lens when it is set at 8 mm or at 15 mm for the focal length and when simultaneously the focue is respectively set at the closest distanec or at the infinty. I have build an overview of the result computed by Canon:

The image radial mapping is considerably different when the distance of focus is set at 150 mm (i.e. closest available distance of focus) than when it is set at infinity. Beware: these charts drawn by Canon are not taking into account the restriction due to possible cropping by the sensor limited coverage (even on a FF camera): on an actual fulframe camera, the FOV for the longer focal end (i.e14 mm. < L < 15 mm) may be smaller than the value of ω written on the chart.

Expected resolution of the equirectangular panorama output

The range of focal length of the Canon zoom fisheye encompasses probably most of the needs of the panorama photographer when the rendered output is to be viewed on the screen of a computer or a mobile device. Only very high resolution panoramas (often being very large mosaics of images presented in cylindrical or Panini projection) would required a different lens.

Mainly using the resources of the above linked Images Databank, I have collected the data that were computed by the AutoPano Giga and PTGui Pro software when asking to "Create or Render the panorama": the Maximum (No loss of detail) of the possible Optimum size has been selected and the proposed resulting sizes were listed on a table. I admit that these numbers could have been rounded...

Lens model and Focal length
Shooting scheme
Resolution (Optimal & Maximal size in Pixels)
AutoPano Giga
PTGui Pro
Fisheye (no sub-option: => Circular)
Option "Circular"
Option "Full Frame"
Canon 8-15 mm @ 8 mm
3 Horizontal
8320 x 4160
7172 x 3586
7172 x 3586
Canon 8-15 mm @ 8 mm
4 Horizontal
8062 x 4031
7184 x 3592
7172 x 3586
Canon 8-15 mm @ 9 mm
3 Horizontal
8996 x 4498
7916 x 3958
8060 x 4030
Canon 8-15 mm @ 10 mm
3 Horizontal
10234 x 5117
8944 x 4472
9304 x 4652
Canon 8-15 mm @ 11 mm
3 Horizontal
11152 x 5576
9736 x 4868
10276 x 5138
Canon 8-15 mm @ 12 mm
4 Horizontal
10772 x 5396
10572 x 5286
11296 x 5648
Canon 8-15 mm @ 13 mm
Nadir + 4H + Zenith
13906 x 6953
11608 x 5804
12388 x 6194
Canon 8-15 mm @ 14 mm
Nadir + 6 Horizontal + Zenith
13908 x 6954
12572 x 6286
13468 x 6734
Canon 8-15 mm @ 15 mm
Nadir + 4 Horizontal + Zenith
15058 x 7529
14036 x 7018
14816 x 7408
Canon 8-15 mm @ 15 mm
Nadir + 6 Horizontal + Zenith
15154 x 7577
14004 x 7002
14788 x 7394

BTW there are some obvious discrepancies in the results from Kolor's APG: the results above were obtained by using the latest stable official APG version (v 2.5.2). A completely overhauled version including new math models for the lenses (and/or for correction of the distortion) should be released soon (Beta testing is still under way). The results of optimum size computation should then be more accurate and homogenous.

These data can be compared with the size got by stitching images from other fisheyes and from some rectilinear lenses in an other article.


About Focal Lengths and about Range of focal length

The radial mapping chart that is presented in the above paragraph shows the lower end of the range behavior of the Canon zoom being "almost" identical to the Sigma 8 mm mapping near the center of the image. Note that's here (on that chart) the newest (f3.5) Sigma model. Therefore one can assume that the shortest focal length of the Canon zoom is probably 8 mm and this can be verified by a simple trigonotric computation.

The upper end of the range of focal lengths is said by Canon to be 15 mm. However, by comparing images by the EF 8-15 mm @ 15 mm with that by venerable 15 mm f2.8 (also by Canon) and additionnally with some selected focal lengths of the Tokina10-17 mm (i.e. 15 mm and 16 mm respectively) it seems that Canon has designed a 16 mm fisheye lens. The estimation given by PTGui goes for 15.5 mm to... 15.95 mm. Would 8-16 mm be a more exact designation?. This seems however not to be completely confirmed by peering at the edges of the real photographs that were shot to compare Image Quality (later in this article). As a matter of fact, these images tend to prove the accuracy of the Canon designation of 15 mm for the maximum focal length. The absence of positive stops to accurately set the focal length (15, 16 mm...) on the Tokina may have contributed to the possible confusion.

Side note:

The meaning of << focal length >> of a fisheye is the subject of controversy. It depends on the user's context. For a standard and ideal (rectilinear) lens and limting the study to paraxial Optics, the initial signification linked to the inclination of the tangent to the curves at coordinate (0,0) is fine and unambiguous. This definition of the focal length had in this simplified model a tangible sense for standard undistorted lenses and it can even then be related to the angle of view by a simple mathematical trigonometric function.

However these simplified notions of focal length have been comprehended by the public in a way that can be summarized by a simple saying: "the shorter focal length yields the wider field". Many photographers and many reviewers makes this mental assimilation. The recent introduction of small digital sensor has further accentuated the assimilation by adding the questionable "equivalent focal length" to the list of spec... Today it is nearly totally incongruous to write the words "Focal length" in a sense and in a context that does not implies "Angle of view". The focal length is assumed to be constant over the whole field... Alas! this assumption generally may not be valid at all with (real) lenses that are affected with distortion of any kind: together with the distortion, the focal length may be described as a function of the incident angle and it thus varies in the field (DxO Labs literature). For a fisheye lens this may lead to bizarre graphs where the focal can be zéro and/or distortion can reach infinity on assymptote! While mathematically correct this is practical nonsense especially when the its author insists in using Gaussian optical theory and (undistorted) Standard (rectilinear) lenses for Unique Reference. In wich comprehensible way can a human mind can comprehend what is an infinite distortion?

In short, and to be realistic and to take into account the actual casual use of some of the Optics vocabulary (especially on the Internet), one should not hang to a rigid mathematical definiton. Almost any number that is designated for description of the focal length by the manufacturer of a fisheye lens can be and shall eventually be disputed. And sometimes this is done in a large extent. This has happened and puzzled the customers base when the Samyang 8 mm fisheye was released some months ago: the same lens is sold with a focal length designation ranging literally from 6.5 to 8mm by different re-branding depending apparently only on merchandising law to post a "different label" from the others.

BTW one can again observe that the Samyang 8 mm mapping (representing the projection) looks really different and peculiar on the PDF chart: its curve is the only one that flexes upward when the angle is increasing! That's the direct effect of the stereographic projection!


Angle of view for the different focal lengths: is 2-shot panorama feasible with the new Canon 8 mm fisheye lens?

The Angle of view of the lens on a FulFrame sensor looks very familiar and matches those of similar lenses with about 188 degrees (for cropped circular image). Nevertheless, the Angle of View is severely restricted by hard optical vignetting in the lower part of the focal length range (below ~10 mm). At 8 mm the full circular image doesn't contain useful pixels coming from more than 90 degrees from the axis when the aperture has been shut down to more than f/8. By "useful pixel" I mean those that are really clean from bluish seam. This is IMO sad news and great disappointment to this author. I had been hoping that Canon could have responded to this plea that I posted some years ago:-(

While it is perfectly possible to stitch a 2-shot panorama, I have found impossible to get rid of the seam (it's generally bluish) at the joint between the two hemispheric images. The IQ of the Canon @ 8 mm near the edges is very good (equal or better than for other 8 mm fisheye lenses) but the circle limit is just 3 to 4 degrees to close to the center... frustration{:1_336:}

Note: Setting the zoom at more than 9 mm (and up to12 mm) and shooting two back-to-back photos in landscape mode: that gives the ability to stitch 360° cylindrical panoramas. Spherical VR are also possible to be stitched in this simplified way, but two holes are left apparent at the nadir and at the zenith. In aerial panorama photography, setting the inclination down at say 20° would fill the hole at the nadir (a further shot of the zenith can possibly be added later to complete the whole sphere and to hide the flying vehicle on the bottom of which the camera is mounted on). BTW this can also be done with the Tokina 10-17 mm (@ 10 to 12 mm) and with the Nikkor 10.5 mm.


The lens cap that loosely fits on the sun hood

The retaining device that should maintain the front protective cap by holding it locked by means of two tongues that grip on the mini-grooved lips on the inner side the sun-hood. That's a common feature that is used by many lens makers. But Canon have based their version for the EF 8-15 mm on a plastic made, elastic ring that is supposed to act as a spring firmly pushing on the two tongues. Actually a simple brushing with a finger tip against one of the two buttons and the cap is eventually released... The elastic ring is so poorly dimensioned that it make the cap more prone to fall on the floor than the cap that fits on the venerable EF 15 mm! The users of that venerable prime fisheye lens shall know what I am referring to:-)

I have made two simple metallic blades to make the spring force much stronger. The cap cannot now unwillingly get loose.

two additional spring blades

now the cap won't get loose unwillingly !

The lens hood

Molded from black plastic material, it is removable (by holding down a small button to release the hood baïonette). After a full month of use, I have never yet shot a photograph with the hood left attached on the lens.... though it is a nice and effective protection of the front lens surface from scratches:-).

BTW will someone design and manufacture a simple removable "cap" that would replace the bulky (hood + cap) present assembly? This is obviously feasible and it would save a lot of space in the congested photo bag of those who do not intend to ever use the current removable sun hood when shooting photos....

The Zoom Limiter Switch is not a click-stop or a zoom locking device{:1_336:}

The seasoned panorama photographer shall be disconcerted by this rather strange invention designed by Canon. I personnally have thought at first to be a useless and silly feature, but after a while, I think now that it could be very useful if it was modified and improved as it could become a simple accurate zoom setting feature with e.g. click-stops or any other means to set repeatable and accurate focal length selection. A zoom locking feature would also be a nice add-on...

At present it is possible to do that only when setting the focal length at 8 mm and 15 mm (against mechanical end-stops) and also 10 mm (by using the zoom limitter switch). I personally would love Canon to extend the 10 mm hard limit to... 12 mm.

Many users shall miss an ability to positively mechanically stop the zoom at some specific main positions where we could set the focal length at some specific precise values on a repeatable way so as to use a template for batch stitching without CP optimization. IMO this should not be very hard to implement.

  Image Quality: comparing with other rival lenses at medium aperture setting (f/5.6 or f/8)

I have cropped a small area (e.g. 450 x 300 px) just near the corner of the full-scale images (or on the edge of the circle where applicable).

To better peer at the details, I have found that zooming in the cropped image in a graphic editing program (to 300% for instance) strengthens considerably the ability to discriminate the best lens from a lesser performer.

I do not own the venerable prime 15 mm model from Canon. It has been rumored to be totally discontinued by Canon since the EF 8-18 mm introduction in end of July 20011. As a matter of fact it disappeared from the store at Canon Europe even if it is still listed on the catalog (Sept 2011) of the Canon USA and Canon Asia South-East web sites. I had however shot some test photographs with a 15 mm lens that Gilles Vidal lent me early in 2007. I subsequently compared one (shot at f/8) with a recent photos also shot at f/8 with the very same EOS 5D FF (12 Mpix) camera with the new Fisheye zoom that was used four years and a half ago. The camera being different from the others cases, I shall present this case separately from the other..

1- Canon 8-15 mm compared with Canon EF 15 mm f2.8 fisheye (EOS 5D = 2912 x 4368 pixels)

Note: Such a comparison has been made already by for instance.

The fixed focal lens was the only fisheye lens that was sold by Canon before they launched the new zoom lens. The reputation of this lens for clarity, sharpness, contrast and the moderate chromatic aberration (easy to completely correct) was well established. With a moderate retail price and as it is fully compatible with the automatic features of the EOS camera family, it was therefore probably the favorite lens used by Canonists panorama photographers to get high resolution on a 180 x 360 panorama. I personally preferred to use the Samyang 14 mm rectilinear lens for this task despite the higher minimum number of individual images required to be shot (8 and 6 respectively)


EF 15 mm f2.8 (Prime lens) @ f/8 EF 8-15 mm f4L USM (Zoom lens) @ 15m & f/8
100% scaled cropped from original
100% scaled cropped from original
reduced from original size (2912 x 4368 pixels) reduced from original size (2912 x 4368 pixels)

There is no visible difference or sharpness or contrast at the center of the pictures, but as can be seen on the example above, the new lens is obviously much sharper at the edge of the frame and especially near the corners.

2- Canon 8-15 mm compared with Tokina 10-17 mm zoom (EOS 5D Mk2 = 5616 x 3744 pixels)

The Tokina zoom lens is based on an original design by Asahi-Pentax. As a matter of fact Asahi has been granted several patents for fisheye zoom lenses. These patents were supposed to be applicable to retro-focus lenses mounted on a 35 mm film camera of the mid 90's era. Ironically, Pentax has yet to release a full frame (i.e. 24 x 36 mm) digital camera. So they have designed a "down-scaled" version of their older film era model and this has become the Tokina 10-17 mm that is fully compatible with many APS-C DSLR by Canon, Nikon, Sony and others. A Pentax compatible version is also available and is sold under a Pentax (or Samsung) label. The fixed metallic integrated sun-hood that restricts the field of view can be "shaved" in order to fit perfectly on FF DSLR camera; Hence some Tokina photos presented hereafter.

@ 15 mm f/5.6

Tokina 10-17 mm @ 15 mm @ f/5.6

Canon 8-15 mm @ 15 mm @ f/5.6

The Tokina lens is known to be a bit soft on the peripheral part of the image at the longer end of the focal length range. This is confirmed here. No surprise: the Canon lens surpasses the older Pentax designed lens and it's obvious on this picture samples.

@ 12 mm f/5.6

Tokina 10-17 mm @ 12 mm @ f/5.6

Canon 8-15 mm @ 12 mm @ f/5.6

It seems that the image from the Canon lens is just a little crispier than its counterpart from Tokina. Both are excellent though.

  2- Canon 8-15 mm compared with Tokina 10-17 mm zoom @ 10 mm and with Nikkor 10.5 mm (EOS 5D Mk2 = 5616 x 3744 pixels) @ f/5.6
Tokina 10-17 mm @ 10 mm @ f/5.6 Nikkor 10.5 mm @ f/5.6 Canon 8-15 mm @ 10 mm @ f/5.6

Exept very near the limit of the circle where the Nikon lens becomes soft (and excessively compressed) all lenses are very sharp. Contrast is obviously lower and unfortunatly some residual chromatic aberration cannot be corrected when the Nikon and the Tokina are concerned.

The image IQ is apparently the same on the important major part of the circular image (that shall still be visible after stitching and blending) on the outputted panorama image. As the peripheral inner part of the circle is not really included into the final panorama, it would be hard to decide in practice what is the sharpest lens between all.

Due to the cleanest overall appearence of the Canon image after CA correction, I tend to prefer the 8-15 mm lens before the two others. Performance at other aperture settings could also be compared in a possible tie-breaker way...

  3- Canon 8-15 mm compared with Samyang 8 mm and with Sigma 8 mm f4.0 (EOS 5D Mk2 = 5616 x 3744 pixels) @ f/5.6
Samyang 8 mm @ f/5.6 Sigma 8 mm f4.0 @ f/5.6 Canon 8-15 mm @ 8 mm @ f/5.6

The Sigma lens is obviously the worst performer of all. This lens is clearly outdated (and discontinued in 2006) and it has been replaced by an improved (and of completely dofferent design) model with a wider maximum aperture (Sigma 8 mm f3.5).

The Samyang is of a completely different projection type (Proportional aka Stereographic fisheye). The outer part of the circular image is much less compressed than it is in the other commun cases of fisheye lenses. The resultant effective "magnification" in this region makes the image to appear "softer" but reveals its quality in comparison when magnification is visually equalized: it's then a bit sharper than the Canon lens image.

The image from the Canon lens is anyhow excellent at this focal length.


Canon 8-15 mm Vs Tokina 10-17 mm : two zoom lenses compared at fully opened aperture (f/4).

This stop setting is the widest possible on the Canon lens and on the upper end of the focal length range of the Tokina zoom lens. To better assess the possible IQ, the images have been corrected from CA and light fall-off after proper cropping in ACR5.6 and in contrast with the shooting process that was used in the previous paragraph, the camera has been pivoted by 90 degrees between two shots. The target object stays 1.8 meters (i.e. the hyperfocal @ 15 mm) distant from the sensor during the shooting sequence.

You may view and compare the resulting images at 15 mm, 12 mm and 10 mm on an other page and on a table here.


Chromatic aberration

A moderate level of Chromatic aberration is visible on the images shot with a Canon 8-15 mm lens. Some reviewer have described this aberration as large or very large. At first glance, the amount is less or about the same as on images of most similar and rival fisheye lenses. The exceptional sharpness that extends far from the center of the image and for the whole focal length range of the Canon zoom lens makes in fact the CA to look less pregnant than for other softer similar fisheye lenses, yet it may -as a consequence- strengthen the color saturation of the dreadly fringe and thus it may impress the " CA uneducated" tester.

Unlike for most of the other fisheye lenses cited in this article, Longitudinal Chromatic Aberration is probably absent (or it is in practice negligible). Consequently the correction of the sole Lateral (aka Transverse) Chromatic Aberration is actually feasible all over the whole image (from center to edges) on a reproducible maner and the result is thus more gratifying. This is not the case for example with the Nikkor 10.5 mm and with the Tokina 10-17 mm zoom were a compromise has to be applied in order to leave some level of residual CA near the edges of the circle so as to fully get rid of it on the most important parts around the center of the image. One can observe this on the small cropped images in the above paragraph.

More surprizing from the new Canon lens is the ability by the panoramic photographer to correct the chromatic aberration with a sole setting in his raw converter (e.g. ACR 5 or 6) for the whole focal length range! Only one set of Red/Cyan and Blue/Yellow correction figures for a perfect correction!

Like for all other lens, purple fringing is also present. I suspect that most if its presence is caused by the camera itself (this is disputed by some authors though). I have however found that it can be better corrected on images shot with the new Canon lens. I must reckognize this as being a subjective observation. I would appreciate to know the reader's opinion on this topic...

Recall: you may experiment with some images from the Canon new fisheye zoom lens after download of RAW (.CR2) files from a Test Images Databank.

Ghosting and flare

I was rather sceptical one year ago (on 26 August,2010) when reading the news on "Canon releases EF 8-15mm f/4 L USM fisheye zoom lens":

<< Ghosting and flare caused by strong light sources can be a particular challenge when shooting at such a wide angle, and, in order to counteract these artifacts, all lens elements have been covered with Canon’s Super Spectra Coating. The inner surface of the front element also features Canon’s SubWavelength Structure Coating (SWC), which uses tiny cone-like structures, shorter than the wavelength of visible light on the surface of the lens element, to gradually slow light down until it is at the same speed as it travels through glass, removing the risk of flare occurring before the light reaches the sensor.>>

This sounded like a pure commercial advertisement. I must admit now that the text that had written a year ago without ever having had the lens in their hands, appears in fact to tell nearly the truth. After more than one month of use and stitching hundreds of panoramas I could not find one occurrence where colored spots of flare could be obviously apparent and annoying, whatever the lighting conditions. Veiling glare near strong lights (e.g. the sun) will of course reduce the contrast (sometimes severely). Some veiling shall happen when shooting in front of a strong source of light such as a large window, but this is considerably less pregnant for this lens than with all the others that I use for panorama photography. had posted a page that shows some artifacts especially on a photograph taken at 12 mm with a very faint and pale ghost visible on the pavement. This is amazing: all of my "similar" photographs that I shot at the same aperture setting (f/8) never show even a visible trace of flare ghost.

So... I have studied a bit more closely the subject by putting the lens to a test series that should reveal the colored ghosts.

Before performing the test, I knew (or believed) that

  1. The colored artifacts due to flare are generally more visible against a black background: I have thus shot within a somewhat "dark" room or in the night aiming toward the sky.
  2. The colored artifacts due to flare are more visible when the image of the strong light source is over-exposed.
  3. The colored artifacts due to flare are only visible when the strong light source is coming from the edge of the circular coverage or close to that edge. Let's remove the sun hood to favor the flare happening.

I thus used the same LED lighting set-up that was already used successfully to compare some other fisheyes to first shoot in the night toward the sky and got an idea of what to expect (at the two ends of the focal length range):

8-15 mm @ 8 mm
8-15 mm @ 15 mm
Note: The lens front surface was cleaned with a cloth... "as usual"
Note: "Extra care" was used to clean the lens front surface

Click on the picture to enlarge the image.

There is a bright with rainbow ed colors artifact on the circular image @ f=8mm (see red arrow). The color saturation and intensity decrease rapidly when the focal length is increased. It is just barely visible when f > 9.5 mm and invisible when f > 12 mm.

The flare ghosting artifacts that are very small indeed, happen to be visible only at the very low end of the focal range! With a full frame camera, I almost never shoot below 12 mm of focal length. That is why I did not yet see these colored dots on my panoramas!

Lets now see whet happens with night/day photos stil using the LED in the corner of the frame:

8-15 mm @ 8 mm (Night)
8-15 mm @ 8 mm (Day)
The ghosts are again there, but their number is smaller than is visible in complete darkness of the night...
Some ghosts are still visible (red arrows) but only few of them and they are tiny dots...

As a matter of fact, the small artifacts may often not be visible on the panorama because of the tiny size of them or they may become invisible after blending it with another (ghost-free) contiguous image...


In natural outdoor and sun-lit scenes and in specific shooting configuration, the flare artifacts may be visible against a cloud-free sky:

Click on the picture to see the red circled area in a larger scale...

The size of the dreaded dots is small and similar to the artifact that I can see on images from other lenses. That's is generally very easy to clone out with a graphic editing software...

The large colored ghosts caused by the sun and that sometimes spoil the lower part of the image when using the Sigma8 mm, the Samyang 14 mm and in a lesser extent the Samyang 8 mm seems not to be possible to plague the images from the new Canon lens.

Compared to all the fisheye lenses and UWA standard lenses that I have tested or used, the new Canon zoom lens is most probably the best to not be prone to show flare artifacts.

BTW I am still studying another phenomenon that I have spotted once and that probably is a consequence of internal reflections. The problem is to successfully replicate this phenomenon! Please stay tuned: this paragraph shall be completed soon.


Overview and Conclusion

Canon has recurrently bragged that the 8-15 mm zoom range is ideal to suit all the possible requirements for all Canon DSLR users (wanting a fisheye lens) whatever the size of the sensor of the camera. I agree, but I was very frustrated to measure not more than 180 degrees of Angle of View when the focal length is set at 8 mm.

Because of the meager angle of view at the low end of the focal length range of the zoom and because I use almost exclusively a full frame DSLR, I shall certainly very rarely shot with a zoom setting below 12 mm or so and probably never below 10 mm except for some rare test shots.

Designing a prime 8 mm (or less) lens with a larger than 185 degrees of Angle of View is certainly feasible. I must admit however that doing so for a ZOOM lens which range encompasses Fullframe is a real challeng. The excellent general pertinence and balance of the present zoom concept would have been shattered if Canon had tried to enlarge the Angle of view of the un-cropped circular (hemispherical) image on a FF camera: I realize that in order to possibly get a much wider FOV, they would have had to extend considerably the diameter of some glass elements... including the frontal one. Bulkiness, weight and cost (price) would probably have been out of commercially acceptable range.

Despite this IMO negative discovery, I have found that this is the best and more versatile lens for panorama photography that I could put my hands on. The NPP/LPP shift when zooming in (or out) is in practice negligible and the fisheye projection type is rather classical (Equi-Solid angle projection close to being Equidistant) is easily treated by the stiching software.

When it is correctly set to hyperfocal, it is sharp over the whole image and possibly the sharpest of all presently available fisheye lenses for panorama photography. It makes images with incredible high contrast and it is almost immune to flare ghosting. The Transverse ChromaticAberration is homogenous and linearly spread: it can perfectly be corrected by RAW converting software. In all these IQ domains, the zoom lens equals or even betters the sharpest UWA lens that I had in my stable, yet it demands fewer shots to get a full spherical panorama at nearly the same output natural resolution.

There are few minor quirks. The silly lens cap that does not hold strongly enough on the sun hood is an example of poor design. The engraved scale under a plastic transparent window for setting the distance of focus is another flawed feature without to much consequence for panorama shooting though. The funny Zoom limiter switch can be useful to accurately set the zoom at three discrete values (8 mm, 10 mm and 15 mm) but there is no way to set at say 12 mm in a reproducible maner and with precision. (Update note * appended below)

I shall use it for single-row (@12 mm) and/or high resolution (@15 mm) full spherical panorama. In other words, the Canon 8-15 mm f4 "L" as it is now, shall be the sole lens for normal panorama photography to put in the bag from now on. My bag shall therefore be lighten but my wallet has also become lighter because of the steep price that I had to pay for possession and having the pleasure to use it;-)


Michel Thoby

Issue 1: 5 September 2011.

( last Rev: April 2012)

评论 (0 个评论)


您需要登录后才可以评论 登录 | 立即注册