I recently started using MTF Mapper.
I'm very impressed with this product, which allows for stable, quantitative
data acquisition with simple settings.
I'm grateful for this product because it allows everyone to accumulate data
by setting the same conditions.
Thank you.
I have a few questions after using it, and I would appreciate your advice.
Is it possible to set the scale to a fixed value to make it easier to
compare lens performance?
I would particularly like to set the scale of the chromatic aberration
graph and grid2D graph to a fixed value.
I've figured out how to set the scale of grid3D to a fixed value, so
that's not a problem.
Is it possible to correctly analyze pixel-shifted images by simply
halving the pixel size?
I halved the pixel size in MTF Mapper and analyzed the pixel-shifted
images.
The results were slightly better than normal shooting, but I'm not sure
if these results are correct.
Is it possible to increase the number of lens profiles to more than
three?
When comparing low-frequency and high-frequency performance, if there
are only three lens profiles, you have to compare them twice.
I would like to be able to plot around six if possible.
Is it possible to increase the number of MTF-XX plots to more than two?
I prefer MTF at 30%, but 50% is the standard.
It would be very useful to be able to generate both at the same time.
Is it possible to change the naming convention for the graphs?
Currently, the graph name is added to the end of the filename, but I
would like to add it to the beginning (this will make it easier to compare
different lenses as they are sorted consecutively).
The actual test setup is as follows:
Magnification: x0.1 (about A4 size)
Camera: Z8
The actual data and settings I created have been uploaded below.
It sounds like you already have solutions to some of your questions, but I'll try to address them:
* Adding a way to fix the scale for chromatic aberration or the 2D grid plots via the GUI is possible, and something that would not take a huge amount of work, so I might consider it;
* I don't have access to any cameras to produce pixel-shifted images, so I'm hesitant to implement any corrections. But remember that pixel shifting increases the sample density only, it does not alter the low-pass filter effect of the lens, nor does it change the size of the photosite over which light is collected. I know that photosite pitch and pixel size are often used interchangeably, but with pixel shifting you are not changing the pixel size at all, although one might argue that you are affecting the effective photosite pitch. So from my perspective the actual system MTF would not change at all when using pixel shifting. Changing the photosite pitch (incorrectly called the "pixel size" in the MTF Mapper GUI) should indeed allow MTF Mapper to represent the spatial frequencies correctly, but the results should be almost identical to the normal, non-shifted MTF if you plot them side-by-side after using the correct pixel pitch for each image. Of course "developing" the pixel-shifted images could add processing steps like more aggressive sharpening to take advantage of the reduced aliasing in the pixel-shifted image (the true benefit of taking pixel-shifted images), so you would need the equivalent of dcraw or libRaw with pixel-shift support to know for sure.
* Hmm. While it is not hard to increase the number of lens profile frequencies, it seems like a very niche feature. It sounds like you already figured out how to do this on the command line, which is what I would recommend.
* The existing output formats, e.g., "edge_mtf_values.txt" do not lend themselves to representing multiple MTF-XX values in a single file. The easy way is to run MTF Mapper twice with different --mtf options, but if you really need to cut down processing time by running MTF Mapper only once you can extract arbitrary MTF-XX values directly from the "edge_sfr_values.txt" by interpolating the SFR.
* Ok, it's not that hard to change the file name convention so that the type becomes a prefix, but doing so would break compatibility with hundreds of other users. It sounds like you have already found a workaround, though.
Judging from your second post, it sounds like you are already processing from the command line, which makes a lot of the requests above fairly straightforward if you ask chatGPT for some gnuplot tips. But in case you have any specific questions regarding command-line usage, please feel free to ask!
Regards,
Frans
❤️
1
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Thank you for all your advice.
I was just writing various scripts this morning to batch process a large amount of data.
I learned a lot by comparing various MTF output results and line-of-resolution data.
It's very exciting and fun.
I was also able to confirm that the qualitative evaluation based on the analysis results and the subjective quantitative evaluation results tend to coincide.
I'm really looking forward to comparing the many lenses I have at hand.
Thank you for creating and providing such a wonderful tool.
*I'll look into the gnuplot commands for the graph later.
*Regarding pixel shift
I converted the image to tiff and processed it in Photoshop, and found that the line-of-resolution improved at very low MTFs, such as MTF 10.
If you think about it, low frequencies are resolved even in images before pixel shift, so pixel shifting doesn't necessarily increase the line-of-resolution.
The analysis results seem to improve at values above 100 lines/mm.
As you said, there doesn't seem to be anything like dcraw or libRaw that supports pixel shift, so right now I can only develop images without adding sharpness correction.
If I could output a linear Tiff from RAW, I could split it into R, G, and B channels and analyze each as grayscale data.
Since pixel shift data should have all the RGB data, it seems easier to handle than Bayer images, which require demosaicing.
I'll continue to look into this area.
*About the sample
I've uploaded the results of various analyses and pixel shifts below. https://drive.google.com/drive/folders/1yJYy4i9in9tUOcb-VPESqCBVv6Er1Jkw?usp=drive_link
I'm not sure if they'll be of any use, but I think it'll be fun to compare them.
The pixel shift sample data also includes data from a photograph of a chart for qualitative evaluation (called LINE).
From this image, it appears that resolution is barely achieved up to around 200 lines/mm.
Thank you for this wonderful tool!
It's a shame that donations can't be made from my country.
I look forward to your continued development.
Gray-Box
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
I'm sorry to ask a question after I've already done so, but I've read the help and understood a lot of things.
What I wanted to do was use command line mode and I was able to automate some of the tasks by renaming on the command line.
I'm going to keep using command line mode for a while.
I'll ask again if there's anything I can't solve on the command line.
Thank you.
Gray-Box
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Dear Author,
I recently started using MTF Mapper.
I'm very impressed with this product, which allows for stable, quantitative
data acquisition with simple settings.
I'm grateful for this product because it allows everyone to accumulate data
by setting the same conditions.
Thank you.
I have a few questions after using it, and I would appreciate your advice.
compare lens performance?
I would particularly like to set the scale of the chromatic aberration
graph and grid2D graph to a fixed value.
I've figured out how to set the scale of grid3D to a fixed value, so
that's not a problem.
halving the pixel size?
I halved the pixel size in MTF Mapper and analyzed the pixel-shifted
images.
The results were slightly better than normal shooting, but I'm not sure
if these results are correct.
three?
When comparing low-frequency and high-frequency performance, if there
are only three lens profiles, you have to compare them twice.
I would like to be able to plot around six if possible.
I prefer MTF at 30%, but 50% is the standard.
It would be very useful to be able to generate both at the same time.
Currently, the graph name is added to the end of the filename, but I
would like to add it to the beginning (this will make it easier to compare
different lenses as they are sorted consecutively).
The actual test setup is as follows:
The actual data and settings I created have been uploaded below.
-
https://drive.google.com/drive/folders/1t5Jal-Sun8_pgm_Sbtz2uDSwliC437vL?usp=sharing
I would like to fine-tune the settings before testing various lenses.
I look forward to your comments.
Gray-Box
Hi Gray-Box,
It sounds like you already have solutions to some of your questions, but I'll try to address them:
* Adding a way to fix the scale for chromatic aberration or the 2D grid plots via the GUI is possible, and something that would not take a huge amount of work, so I might consider it;
* I don't have access to any cameras to produce pixel-shifted images, so I'm hesitant to implement any corrections. But remember that pixel shifting increases the sample density only, it does not alter the low-pass filter effect of the lens, nor does it change the size of the photosite over which light is collected. I know that photosite pitch and pixel size are often used interchangeably, but with pixel shifting you are not changing the pixel size at all, although one might argue that you are affecting the effective photosite pitch. So from my perspective the actual system MTF would not change at all when using pixel shifting. Changing the photosite pitch (incorrectly called the "pixel size" in the MTF Mapper GUI) should indeed allow MTF Mapper to represent the spatial frequencies correctly, but the results should be almost identical to the normal, non-shifted MTF if you plot them side-by-side after using the correct pixel pitch for each image. Of course "developing" the pixel-shifted images could add processing steps like more aggressive sharpening to take advantage of the reduced aliasing in the pixel-shifted image (the true benefit of taking pixel-shifted images), so you would need the equivalent of dcraw or libRaw with pixel-shift support to know for sure.
* Hmm. While it is not hard to increase the number of lens profile frequencies, it seems like a very niche feature. It sounds like you already figured out how to do this on the command line, which is what I would recommend.
* The existing output formats, e.g., "edge_mtf_values.txt" do not lend themselves to representing multiple MTF-XX values in a single file. The easy way is to run MTF Mapper twice with different
--mtfoptions, but if you really need to cut down processing time by running MTF Mapper only once you can extract arbitrary MTF-XX values directly from the "edge_sfr_values.txt" by interpolating the SFR.* Ok, it's not that hard to change the file name convention so that the type becomes a prefix, but doing so would break compatibility with hundreds of other users. It sounds like you have already found a workaround, though.
Judging from your second post, it sounds like you are already processing from the command line, which makes a lot of the requests above fairly straightforward if you ask chatGPT for some gnuplot tips. But in case you have any specific questions regarding command-line usage, please feel free to ask!
Regards,
Frans
Dear Frans,
Thank you for all your advice.
I was just writing various scripts this morning to batch process a large amount of data.
I learned a lot by comparing various MTF output results and line-of-resolution data.
It's very exciting and fun.
I was also able to confirm that the qualitative evaluation based on the analysis results and the subjective quantitative evaluation results tend to coincide.
I'm really looking forward to comparing the many lenses I have at hand.
Thank you for creating and providing such a wonderful tool.
*I'll look into the gnuplot commands for the graph later.
*Regarding pixel shift
I converted the image to tiff and processed it in Photoshop, and found that the line-of-resolution improved at very low MTFs, such as MTF 10.
If you think about it, low frequencies are resolved even in images before pixel shift, so pixel shifting doesn't necessarily increase the line-of-resolution.
The analysis results seem to improve at values above 100 lines/mm.
As you said, there doesn't seem to be anything like dcraw or libRaw that supports pixel shift, so right now I can only develop images without adding sharpness correction.
If I could output a linear Tiff from RAW, I could split it into R, G, and B channels and analyze each as grayscale data.
Since pixel shift data should have all the RGB data, it seems easier to handle than Bayer images, which require demosaicing.
I'll continue to look into this area.
*About the sample
I've uploaded the results of various analyses and pixel shifts below.
https://drive.google.com/drive/folders/1yJYy4i9in9tUOcb-VPESqCBVv6Er1Jkw?usp=drive_link
I'm not sure if they'll be of any use, but I think it'll be fun to compare them.
The pixel shift sample data also includes data from a photograph of a chart for qualitative evaluation (called LINE).
From this image, it appears that resolution is barely achieved up to around 200 lines/mm.
Thank you for this wonderful tool!
It's a shame that donations can't be made from my country.
I look forward to your continued development.
Gray-Box
Dear All,
I'm sorry to ask a question after I've already done so, but I've read the help and understood a lot of things.
What I wanted to do was use command line mode and I was able to automate some of the tasks by renaming on the command line.
I'm going to keep using command line mode for a while.
I'll ask again if there's anything I can't solve on the command line.
Thank you.
Gray-Box