I've been working on making lens profiles for some manual lenses with medium format cameras and have been wondering what EXIF data is needed by ALPC. I believe I may be missing some necessary information, as the vignette model is definitely not accurate when graphed, which also makes me question the accuracy of the distortion model (seems to be ok). What are all of the EXIF tags used by ALPC to create the distortion and vignette models?
ALPC would be looking at camera make, model, lens model, lens ID, sensor format factor (crop factor) to auto-match the lens profiles. For sub-profile matching, ALPC looks at FocalLength, ApertureValue and focus distance info.
Some these are in the standard EXIF tags; while others are parsed from the EXIF private MakerNote. You can open one of those lens profiles in a text editor to see what is inside the file.
I don't think we're on the same page here... I'm asking about the creation of a lens correction profile, not the matching and application of a lens correction profile. I've been using exiftool to supplement necessary EXIF data (e.g. FocalLength, FNumber, FocalLengthin35mmFormat, SubjectDistance, FocalPlaneXResolution, FocalPlaneYResolution) to my images before using them to generate a lens profile in ALPC. Since I've started creating profiles, I've found a couple EXIF data tags that were missing. When I add them to the images and re-generate the lens profile with that same image set, the distortion and vignette parameters change, which proves that they are used by ALPC. I'm starting to wonder how many other ALPC input parameters (EXIF tags) I'm missing. So... Assuming someone has a fully manual system and has to add ALL of the EXIF data that ALPC would need to mathematically calculate accurate distortion and vignette models, what EXIF data would have to be added?
At creation time, ALPC also need the camera make, model and lens model, lens ID to group chart images into sets. These info are also propagated into the resulting lens profile.
As far as impacting the resulting lens profile data, the FocalLength, FNumber are required. The actual SubjectDistance is estimated by the APLC. So that is optional. You need FocalLengthin35mmFormat or (FocalPlaneXResolution,FocalPlaneYResolution,FocalPlaneResolutionUnit ), but you don't need both.
Hope this is helpful,
Even though it seems I have all of the necessary info included in the EXIF data, I still think that something is not quite right with the vignetting model. When I plug the generated coefficients into the equation specified in your documentation, the resulting curve seems to have the correct form, but it is compressed to the point where edge cases for the given lens are obviously not correct. I've specified one parameter, two parameter, and three parameter when creating the lens profile, and all of the resulting coefficients produce a curve that seems intuitive and starts at the correct value, but is definitely not correct near the edge of the lens (the graph is compressed). Could the use of a medium format sensor be causing these problems?
The v.1 of the Adobe Vignetting Model does have known limitation for a combination of large format sensor and wide angle lenses. A later update to the vignetting model (included in LR4/ACR7) supports a much general model that is more accurate for these cases. But we currently have no upadate to ALPC that generates the general model from image based calibrations. The new model is only utilized for lens profiles created from the lens design data.
The other thing that you could try when you calibrate your lenses. Try using a checker chart with smaller squares and frame the checker board image as close to the image frame as possible when you do the shooting (but still not get clipped). This will allow APLC get a better read of the actual light falloff at the edge, intead of just trying to extrapolate based on the data available only at the image center.
Hope this helps,
jasonoats: How are you calculating the vignette curve? One option is to optimize for the edges (e.g., constrain the optimization to match the corner falloff exactly) so that you don't have under/overcorrection at the farthest corner.
The other thing to consider during the application of the profile is the appropriate sensor size (you touched on this issue). ACR/LR needs to know the pixel geometry, i.e., the size of a pixel, so that it can translate physical measurements (e.g., in the lens profile) to pixel measurements for the purposes of applying the correction.