Anyone making lens profiles for PShop CS5?

[soboyle]

I was surprised that the 5DII wasn't well represented in the existing lens profiles provided by adobe at release of CS5.
Has anyone tried the lens profile creator yet?

[Schewe]

I was surprised that the 5DII wasn't well represented in the existing lens profiles provided by adobe at release of CS5.
Has anyone tried the lens profile creator yet?

Just so you understand, it ain't the camera (or sensor) that matters, it's the lens and whether a lens was profiled on a full frame or a crop sensor...all the EFS Canon lenses were done on a 50D I think and all of the standard lenses on a 5D MII or 1Ds MIII. The reason that Adobe has the camera model in the profile was that they were keeping what lens had been shot with what camera. So, it's the lens not the camera that matters except for situations where a user is making their own. If you have both 1.6 crop and full frame cameras and want to profile a lens you use on both, you'll want to shoot the full frame not the crop sensor because you'll need the lens corrections being created based on the largest area...once done for the full frame, the crop sensor will catch the corrections even though tey will be less severe on the 16 crop. Same deal with a 1D MIII as well.

[JeffKohn]

I can see how corrections for geometric distortion would only be related to lens and format size, but what about CA? If you create a lens profile with images from a 5D, will that give you satisfactory results for CA correction on a 5D2?

[Schewe]

I can see how corrections for geometric distortion would only be related to lens and format size, but what about CA? If you create a lens profile with images from a 5D, will that give you satisfactory results for CA correction on a 5D2?

Same dealeo if you think about it...regardless of the defect, as long as you've made the lens profile on the LARGEST sensor you'll be using, smaller sensor's images will be corrected correctly...it doesn't work the other way around...

[JeffKohn]

Same dealeo if you think about it...regardless of the defect, as long as you've made the lens profile on the LARGEST sensor you'll be using, smaller sensor's images will be corrected correctly...it doesn't work the other way around...

Yes but the 5D has much lower pixel density, and might not fully resolve small amounts of CA. I was thinking it might make sense to use the highest-density sensor possible for creating the profile in this case.

[dreed]

Yes but the 5D has much lower pixel density, and might not fully resolve small amounts of CA. I was thinking it might make sense to use the highest-density sensor possible for creating the profile in this case.

I think you'll find that the presence of CA in a lens will be a function of the lens and thus subject to being described by a mathematical equation (radial distance from the centre, etc) that describes the makeup of the glass and thereby the resolution of the sensor is not as relevant as you might imagine. I imagine that the growth of CA on the image towards the corners/edge can be mapped in a similar mechanism to vignetting. Thus I don't necessarily need to know what the degradation is at every pixel but rather I can measure it at fixed points and map a function to it.

Where there are issues with all such software that uses a cookbook approach is that there are often variations from lens to lens, so what is perfect for one may only be only 99% for another or 101% of what's required. So as long as Adobe used more than one lens to build up their database (and thus are not prone to supplying data from one particular skewed datapoint) for each lens,  it should be reasonable.

What concerns me more about profile generation is the quite valid comments from DxO about the needs for even lighting, a perfectly flat surface, etc. It would be of great comfort to know that Adobe went to the same lengths that DxO did to generate their profiles rather than followed their own advice. I doubt that there are very many people who can produce profiles to the same high standards that DxO have and to that end, it will be interesting to see how many of the people who use DxO software today will switch to Adobe.

[madmanchan]

A few things ...

First, lateral CA can vary from lens unit to lens unit (for a given lens model), though for lenses that exhibit a lot of CA, the lens-to-lens variation is typically small relative to the overall CA inherent to the lens design. There can also be small variations due to sensor & lighting variations (effectively spectral response of sensor and spectrum of lighting play a secondary role). These are all reasons why manual adjustment controls are provided. For example, in Camera Raw 6.1 there are the "Correction Amount" sliders which are effectively volume controls. If you find that a profile for a lens slightly undercorrects with your camera, you can dial it up (and save the result as a custom default, if you wish). A bit like having a basic "lens profile editor" built into the app.

Second, as noted in Tom Hogarty's comments: you do not need perfect flat (or even close to flat) lighting for the vignette estimation. What you need is consistent lighting from shot to shot, within an image set. That is, don't use a flashlight to illuminate the left side of the chart for the first shot, then the right side for the second shot! As long as the lighting is pretty consistent from shot to shot, the Adobe Lens Profile Creator ALPC can factor out nonuniformities in lighting across the chart. (Hint: The technology is related to, but not the same as, the automatic vignette correction used in Photoshop's pano stitching -- i.e., Photomerge.)

Remember that you are supplying multiple images (at least 3, preferably 5 to 9) within a set to ALPC. It is true that if you were trying to estimate vignette falloff from a single image (e.g., by taking an image of a plain surface), then yes you'd want that surface/field to be uniformly illuminated. But that's not the case here, so those assumptions / requirements / prerequisites do not apply.

[walter.sk]

Remember that you are supplying multiple images (at least 3, preferably 5 to 9) within a set to ALPC. It is true that if you were trying to estimate vignette falloff from a single image (e.g., by taking an image of a plain surface), then yes you'd want that surface/field to be uniformly illuminated. But that's not the case here, so those assumptions / requirements / prerequisites do not apply.

Thanks for your explanations.  I have a question regarding the multiple shots.  I read Adobe's instructions as well as several threads on how best to use the profiler, and I am now confused about the relationship of the camera to each image.

I assume that the center image of 9 in the frame would be the closest to parallel to the sensor.  For those that would be above or below the center, would it be better to rotate the camera up or down, or elevate the camera keeping it as parallel as possible?  And for those images to the left and right of the center, is it better to rotate the camera or move the camera to the left or right?  Or does it not matter?

[Schewe]

I assume that the center image of 9 in the frame would be the closest to parallel to the sensor.  For those that would be above or below the center, would it be better to rotate the camera up or down, or elevate the camera keeping it as parallel as possible?  And for those images to the left and right of the center, is it better to rotate the camera or move the camera to the left or right?  Or does it not matter?

Yes is DOES matter but ironically, you DO want to tilt the camera up/down and twist it from left to right. You want to keep camera in a fixed central point and tilt and turn the camera to do the upper corner shots and try to get the corner of the target into the corner of the frame without actually cropping into and losing any of the squares...

The reason you do want to tilt and turn the camera rather than move the camera up/down and left/right to try to maintain a flat on relationship is that the ALPC algorithms can actually use the tilt and turn to aid in the distortion determination. It helps to tilt and turn not hurt the results..

The one thing that is pretty darn critical is to make sure the actual calibration target is flat when you shoot it. No ripples or wrinkles and no bowing nor torquing of the target. Adobe suggests mounting on foam core and putting the target in a picture frame (without the glass) so the frame keeps the target flat. When I did my profiles I spray mounted my targets to 1/4 plate glass which is very flat.

[walter.sk]

Yes is DOES matter but ironically, you DO want to tilt the camera up/down and twist it from left to right. You want to keep camera in a fixed central point and tilt and turn the camera to do the upper courser shots and try to get the corner of the target into the corner of the frame without actually cropping into and loosing any of the squares...

The reason you do want to tilt and turn the camera rather than move the camera up/down and left/right to try to maintain a flat on relationship is that the ALPC algorithms can actually use the tilt and turn to aid in the distortion determination. It helps to tilt and turn not hurt the results..

The one thing that is pretty darn critical is to make sure the actual calibration target is flat when you shoot it. No ripples or wrinkles and no bowing nor torquing of the target. Adobe suggests mounting on foam core and putting the target in a picture frame (without the glass) so the frame keeps the target flat. When I did my profiles I spray mounted my targets to 1/4 plate glass which is very flat.

Thanks...

As usual, your reply comes with an explanation that really clarifies the issue.

[keithrsmith]

There is one issue with this.  I have made a profile for one of my lenses - the Oly12-60.  This seems to OK - although how does one really test it?.  I have however,  found that there are two other profiles for the same lens on the on-line database - and all 3 give different results!.  Maybe Adobe should give some advice on how to test these things.  I guess that shooting the target, in all parts of the frame - by moving the target, not moving the camera, and then seeing if the distortion is fixed might be the way.  I took an single image of the target filling the frame . but that is at a diffferent focal distance of course.

Keith

[walter.sk]

There is one issue with this.  I have made a profile for one of my lenses - the Oly12-60.  This seems to OK - although how does one really test it?.  I have however,  found that there are two other profiles for the same lens on the on-line database - and all 3 give different results!.  Maybe Adobe should give some advice on how to test these things.  I guess that shooting the target, in all parts of the frame - by moving the target, not moving the camera, and then seeing if the distortion is fixed might be the way.  I took an single image of the target filling the frame . but that is at a diffferent focal distance of course.

Keith

I've never been a "pixel-peeper," believing firmly in real life photography.  What I would do after downloading a profile (or making my own, for that matter) would be to take several frame-filling shots of architectural subjects (other than brick walls, which are really too uneven).  A door with window panes, the wall of a glass and steel building with vertical and horizontal lines across the frame, etc., would probably be ideal targets and are what are recommended by the people who write PTLens.  It is a low priced lens correction program that grew out of Panorama Tools, the Helmut Deutch labor of love that was one of the most flexible stitching programs.  The PTLens folks solicit images from users and make profiles for the program.

Their program corrects for pincushion, barrel and mustache distortion, and I believe more recent iterations of the program address chromatic and other distortions as well.

Oh, yes: I forgot my main point.  I would then apply my new profile to one of those target images, and just eyeball it.