Exif.NikonLd4.LensID favoured to Exif.Photo.LensModel, even if latter is defined in db and former is not

[blueglyph]

Is there an existing issue for this?

• I checked and did not find my issue in the already reported ones

Describe the bug

I created the lensfun calibration data for a 3rd-party lens, then I added that information in Darktable's share\lensfun\version_1\misc.xml file, but Darktable doesn't automatically recognize the lens when I open NEF raw files shot with that lens.

In the lens correction module, I see "253" as default (unknown) model, but the name corresponding to the model I've added is present in the drop-down list.

So I have either to select it manually for each photograph, or I can add the data in a .exiv2 (Linux) / exiv2.ini (Windows) file in my home folder. In that file, I have to put "253" (the LensIDNumber value) or "0" (the LensID value):

[nikon]
253=35mm f/2

If I convert the NEF into DNG, using the Adobe DNG Converter, Darktable automatically recognizes the model without requiring the .exiv2/exiv2.ini file, and it correctly shows the model in the Image Information panel, too.

I finally discovered that this strange behaviour was due to the way Darktable determines the lens model:

• If it finds the Exif.NikonLd4.LensID (or LensIDNumber, that isn't clear) tag in the metadata, it uses the number as identifier, even if the model is defined in Exif.Photo.LensModel.
• In the NEF, both tags are present, so Darktable takes Exif.NikonLd4.LensID and ignores Exif.Photo.LensModel. However, it doesn't find the numbers of the former, "0" or "253", in the database, so it gives up.
• In the DNG, only Exif.Photo.LensModel is present, and the value is 35mm f/2. That's the model I added in the misc.xml file above, so it's automatically detected.

My problem with that, once the data on this 3rd-party lens in integrated into lensfun database and ultimately in a next version of Darktable, is that no other user will benefit from the detection. They'll either think it's not supported or will have to manually create the .exiv2 file (if they even know about it). More likely, they'll find it in the Voigtländer models and decide to select it manually each time.

Another problem is that it's impossible to distinguish a series of models using "0" and "253", like the 35 mm and 50 mm models from Voigtländer. Only Exif.Photo.LensModel can tell them apart.

=> Why doesn't Darktable falls back on Exif.Photo.LensModel if Exif.NikonLd4.LensID isn't in any database? Unless there's a way to link both data for all the impacted lenses and for everyone (without asking them to add files manually)? Perhaps if that value is "0", or if Exif.NikonLd4.LensIDNumber is "253", then it should look at the Exif.Photo.LensModel value?

XML content of the calibrated lens

    <lens>
        <maker>Voigtländer</maker>
        <model>35mm f/2</model>
        <model lang="en">APO-LANTHAR 35mm F2 Aspherical II</model>
        <mount>Nikon Z</mount>
        <cropfactor>1</cropfactor>
        <calibration>
            <!-- Taken with Nikon Z6III -->
            <distortion model="ptlens" focal="35" a="0.00726" b="-0.01914" c="0.0101"/>
            <tca model="poly3" focal="35.0" vr="1.0000325" vb="0.9999729" />
            <vignetting model="pa" focal="35.0" aperture="2.0" distance="0.27" k1="-1.7274801" k2="1.4821616" k3="-0.5357087" />
            <vignetting model="pa" focal="35.0" aperture="2.0" distance="0.54" k1="-2.1529816" k2="2.1936554" k3="-0.8936036" />
            <vignetting model="pa" focal="35.0" aperture="2.0" distance="1.62" k1="-2.3381563" k2="2.5067040" k3="-1.0497600" />
            <vignetting model="pa" focal="35.0" aperture="2.0" distance="1000" k1="-2.4054764" k2="2.6108750" k3="-1.0973622" />
            <vignetting model="pa" focal="35.0" aperture="2.8" distance="0.27" k1="-0.4931862" k2="-0.3903400" k3="0.2860779" />
            <vignetting model="pa" focal="35.0" aperture="2.8" distance="0.54" k1="-0.5818163" k2="-0.5645652" k3="0.4422908" />
            <vignetting model="pa" focal="35.0" aperture="2.8" distance="1.62" k1="-0.6329567" k2="-0.6647739" k3="0.5480756" />
            <vignetting model="pa" focal="35.0" aperture="2.8" distance="1000" k1="-0.6536585" k2="-0.7188463" k3="0.6043772" />
            <vignetting model="pa" focal="35.0" aperture="4.0" distance="0.27" k1="-0.5839239" k2="0.0348811" k3="0.0656375" />
            <vignetting model="pa" focal="35.0" aperture="4.0" distance="0.54" k1="-0.7391125" k2="0.1372117" k3="0.0234026" />
            <vignetting model="pa" focal="35.0" aperture="4.0" distance="1.62" k1="-0.8256081" k2="0.2102259" k3="-0.0122386" />
            <vignetting model="pa" focal="35.0" aperture="4.0" distance="1000" k1="-0.8589557" k2="0.2228163" k3="-0.0119729" />
            <vignetting model="pa" focal="35.0" aperture="5.6" distance="0.27" k1="-0.5812536" k2="0.0279456" k3="0.0705040" />
            <vignetting model="pa" focal="35.0" aperture="5.6" distance="0.54" k1="-0.7175983" k2="0.0628979" k3="0.0875625" />
            <vignetting model="pa" focal="35.0" aperture="5.6" distance="1.62" k1="-0.7939760" k2="0.0946543" k3="0.0903589" />
            <vignetting model="pa" focal="35.0" aperture="5.6" distance="1000" k1="-0.8303944" k2="0.1121402" k3="0.0904709" />
            <vignetting model="pa" focal="35.0" aperture="16.0" distance="0.27" k1="-0.6030442" k2="0.0509183" k3="0.0586593" />
            <vignetting model="pa" focal="35.0" aperture="16.0" distance="0.54" k1="-0.7165359" k2="0.0224849" k3="0.1161671" />
            <vignetting model="pa" focal="35.0" aperture="16.0" distance="1.62" k1="-0.7985310" k2="0.0598960" k3="0.1186587" />
            <vignetting model="pa" focal="35.0" aperture="16.0" distance="1000" k1="-0.8312505" k2="0.0598966" k3="0.1325040" />
         </calibration>
    </lens>

Sample image

I can't upload it here, so I temporarily hosted it here.

Exiv2 information

DSC_1956.NEF:
Exif.NikonLd4.LensID                         Short       1  0
Exif.Photo.LensModel                         Ascii      65  35mm f/2

DSC_1956.DNG:
Exif.Photo.LensModel                         Ascii       9  35mm f/2

Steps to reproduce

1. Remove any information related to the lens model from the .exiv2 / exiv2.ini file in the home directory

2. Add the information about the lens model into Darktable's share\lensfun\version_1\misc.xml file

3. Load a NEF photo taken with that model lens
   => Darktable fails to recognize the model automatically

4. Convert the NEF to DNG

5. Load the DNG
   => Darktable successfully recognizes the model automatically

6. Extract the exiv2 Exif.NikonLd4.LensID information from the NEF file (or Exif.NikonLd4.LensIDNumber)

7. Add the information into the .exiv2 / exiv2.ini file

8. Refresh the metadata EXIF information in Darktable lighttable mode

9. Try step 3 again=> Darktable successfully recognizes the model automatically

Expected behavior

No response

Logfile | Screenshot | Screencast

No response

Commit

No response

Where did you obtain darktable from?

darktable.org / GitHub release

darktable version

5.0.1

What OS are you using?

Windows

What is the version of your OS?

Windows 10 x64 22H2

Describe your system

No response

Are you using OpenCL GPU in darktable?

Yes

If yes, what is the GPU card and driver?

No response

Please provide additional context if applicable. You can attach files too, but might need to rename to .txt or .zip

No response

[jade-nl]

Confirmed.

I have 2 voightlanders (35/50mm f/2.0 APO-Lanthar aspherical, Z-Mount). Both used to be found and correctly shown in darktable 4.X. This is no longer the case in 5.X

dt 4.2.1:

lens: 35mm f/2
lens: 50mm f/2

dt 5.0.1

lens: 253
lens: 253

Probably not relevant, but just in case: running on Linux/Debian

[blueglyph]

@jade-nl Interesting that you get the same number for both lenses.

I found that data by extracting the EXIF with exiv2:

Exif.NikonLd4.LensIDNumber                   Byte        1  253

That means this value is unreliable. Maybe Voigtländer doesn't provide the same set of metadata as other vendors, or the camera provides those tags even when the model is unknown. Sadly, I don't have other 3rd-party lenses to check that (preferably from other vendors).

[jade-nl]

Yes, I just double checked and the Exif.NikonLd4.LensIDNumber field for both is 253

The Exif.Nikon3.Lens field shows this:

Exif.Nikon3.Lens                              35mm F2
Exif.Nikon3.Lens                              50mm F2

An the Exif.Photo.LensModel (a none nikon specific field) this:

Exif.Photo.LensModel                          35mm f/2
Exif.Photo.LensModel                          50mm f/2

[blueglyph]

Here's a relevant part of the code that someone else had spotted when I discussed that issue on pixls.us:

    else if(Exiv2::testVersion(0,27,4)
            && FIND_EXIF_TAG("Exif.NikonLd4.LensID") && pos->toLong() == 0)
    {
      // Z body w/ FTZ adapter or recent F body (e.g. D780, D6) detected.
      // Prioritize the legacy ID lookup instead of Exif.Photo.LensModel included
      // in the default Exiv2::lensName() search below.
      if(FIND_EXIF_TAG("Exif.NikonLd4.LensIDNumber"))
        _strlcpy_to_utf8(img->exif_lens, sizeof(img->exif_lens), pos, exifData);
    }
    else if((pos = Exiv2::lensName(exifData)) != exifData.end() && pos->size())
    {
      _strlcpy_to_utf8(img->exif_lens, sizeof(img->exif_lens), pos, exifData);
    }

    // Check if an "Unknown Lens (xxx)" was returned by using regex,
    // and retry Exif.Photo.LensModel as last resort
    GMatchInfo *match_info;

    GRegex *regex = g_regex_new("\\(\\d+\\)$",
                        (GRegexCompileFlags) 0,
                        (GRegexMatchFlags) 0,
                        NULL);
    g_regex_match_full(regex,
                       img->exif_lens,
                       -1,
                       0,
                       (GRegexMatchFlags) 0,
                       &match_info,
                       NULL);
    const int match_count = g_match_info_get_match_count(match_info);
    if(match_count > 0
        && FIND_EXIF_TAG("Exif.Photo.LensModel"))
    {
      _strlcpy_to_utf8(img->exif_lens, sizeof(img->exif_lens), pos, exifData);
    }

With the unfortunate use of a macro #define FIND_EXIF_TAG(key) _exif_read_exif_tag(exifData, &pos, key), so the code is a bit of a hack, but it seems to mean

• take "Exif.NikonLd4.LensID" if present
• else take "Exif.NikonLd4.LensIDNumber"
• if what has been found so far has a number between parentheses, take "Exif.Photo.LensModel"

PS: This part definitely needs rewriting; it's ugly. And why does it show the value of "Exif.NikonLd4.LensIDNumber" instead of "1 0", then?

In comparison, Rawtherapee does like this:

        if (!make.compare(0, 5, "NIKON")) {
            if (find_exif_tag("Exif.NikonLd4.LensID")) {
                if (!to_long(pos)) { // No data, look in LensIDNumber.
                    const auto p = pos;
                    if (!find_exif_tag("Exif.NikonLd4.LensIDNumber")) {
                        pos = p; // Tag not found, so reset pos.
                    }
                }
                lens = pos->print(&exif);
                if (lens == std::to_string(to_long(pos))) { // Not known to Exiv2.
                    lens.clear();
                } else {
                    lens = validateUft8(lens);
                }
            }
        }

where find_exif_tag captures pos and exif and modifies pos, so that code is also a bit murky, but it seems to keep "Exif.NikonLd4.LensID" if it finds a value and only looks "Exif.NikonLd4.LensIDNumber" if it doesn't (same idea as before).

That doesn't explain why it recognizes the lens since it doesn't have the link from "1 0". I suppose that "Exif.Photo.LensModel" is taken first and I missed it (I only had a glance at the code). @jade-nl, by the way, I see you reported a similar problem with that software with F-mount lenses on a Z body, so that doesn't seem to be fullproof either.

It's apparently difficult to determine the model, but all I can say is that it doesn't work in this case for Darktable, while it does for Rawtherapee (and Affinity Photo 2, whose code isn't public).

[kmilos]

Firstly, one should be using exiv2 -Pkvt to check both the original value (v) of the tag, and the internally translated (t) one by Exiv2.

Secondly, there are potentially more tags at play

Exif.Nikon3.LensType                          0  (0)
Exif.Nikon3.Lens                              350/10 350/10 200/100 200/100  35mm F2
Exif.Nikon3.LensFStops                        0 1 12 0  0
Exif.NikonLd4.LensIDNumber                    253  253
Exif.NikonLd4.LensFStops                      0  F0.0
Exif.NikonLd4.LensID                          0  0
Exif.Photo.LensSpecification                  350/10 350/10 200/100 200/100  35mm F2
Exif.Photo.LensMake
Exif.Photo.LensModel                          35mm f/2  35mm f/2
Exif.Photo.LensSerialNumber

As mentioned, from the dt code, Exif.NikonLd4.LensID is checked first for Z mount lenses, and then Exif.NikonLd4.LensIDNumber for legacy F mount lenses (note that other 3rd party lenses like Laowa, Sigma, and Tamron do use the native Z Exif.NikonLd4.LensID, so this is just poor implementation by Voigtlander; not to mention that strings like "35mm f/2" are not exactly unambigous).

Finally, one should actually be putting 253=foo in the .exiv2/exiv2.ini for the workaround. (Thank Voigtlander for using the same ID for more than one lens.)

What's perhaps interesting is that the "translated" Exif.NikonLd4.LensIDNumber value should've ideally come back as "Unknown Lens (253)", and then the dt code should've fallen back to Exif.Photo.LensModel a bit further below indeed.

One could try 253=Unknown Lens (253) via .exiv2/exiv2.ini to force this...?

So, good luck further debugging to whoever has an itch to scratch here. Could as well be related to a change in exiv2 (0.27.x vs 0.28.x), not necessarily in dt code only...

[blueglyph]

I don't understand any more.

Now, not only using .exiv2 in my home directory doesn't work any more, but I have to put either "253" or "0". I'm sure that yesterday I did several tests by using / removing that file, and only "1 0" worked, which doesn't make sense (and I did relaunch Darktable each time, but I wasn't resetting the EXIF in lighttable mode).

I'll update the bug description, but I'm baffled. From the Exiv2 description, the exiv2.ini (in Windows) should use the "LensIDNumber" value, so "253". I suppose it also checks "LensID". What a mess.

The conclusion remains the same: Darktable needs to check "Exif.Photo.LensModel" to differentiate the models in this case (but how to determine "this case" is another matter).

Good luck indeed. I agree that Voigtländer could have put more useful metadata.

[kmilos]

The "1 0" you had was nonsense for .exiv2 - the "1" came from the count of values in a tag (i.e. the "c" option in the default -Pkyct output); only the value is relevant obviously.

From the Exiv2 description, the exiv2.ini (in Windows) should use the "LensIDNumber" value, so "253". I suppose it also checks "LensID".

Nope. It only overrides LensIDNumber when it comes to Nikons. LensID is a more recent Z series tag and thankfully behaves more predictably (or at least it seems to so far).

What a mess.

It is indeed. Thank all the camera and lens vendors for cooking up all their custom and inconsistent metadata schemes over the decades.

[kmilos]

The conclusion remains the same: Darktable needs to check "Exif.Photo.LensModel" to differentiate the models in this case

It does. I suggested the workaround via 253=Unknown Lens (253).

[blueglyph]

The "1 0" you had was nonsense for .exiv2 - the "1" came from the count of values in a tag (i.e. the "c" option in the default -Pkyct output); only the value is relevant obviously.

Indeed. I don't know how it worked for me yesterday. It must have been a remnant of something else that wasn't properly refreshed in Darktable. Sorry for that!

Nope. It only overrides LensIDNumber when it comes to Nikons. LensID is a more recent Z series tag and thankfully behaves more predictably (or at least it seems to so far).

I only meant that it's what I saw in the exiv2 manual, though the explanation is pretty vague. I didn't check the code, but I believe you.

Yes, the workaround is a good idea if someone has several different models. I'm trusting you on that; I haven't tested it (and from the code, it should indeed work).

[kmilos]

Finally, I guess one could add these "253" lenses in the Exiv2 internal translation tables for legacy F mount, like they are in Exiftool (253 is FD in hex). But then you would get a result like "Voigtlander APO-Lanthar 35mm F2" prioritized over "35mm f/2" that would then have a knock-on effect on Lensfun and possibly other apps... Damned if you do, damned if you don't sort of thing. 😮

[jade-nl]

@kmilos

The 35mm f/2 and 50mm f/2 strings might be unambiguous but they were a string that could be used to trigger rules (dt 4.x), at the moment (dt 5.x) there is no way to do this.

This issue is mainly caused by the info that Voigtländer (and probably other third party lenses) do not provide for their lenses. I personally wouldn't mind if darktable does not want to use unambiguous info to determine something, but I do want the possibility to be able to trigger a self made rule (I certainly know what lens I'm using, no guesswork there).

For me it isn't "just" lensfun and its data, I also have individual body/lens profiles made for the white balance module (true 6502K camera references entries) and no way to auto-trigger them in 5.X.

This is, all in all, a nice issue.

[kmilos]

I don't think much has changed re Nikon lens detection between dt 4.8 and 5.0, so someone will have to actually pinpoint to what the problem actually is here, if there is one specific to dt.

[jade-nl]

Having the possibility to use a "free/open" field and its trigger in the rule making section would make things much easier (and is not limited to this specific issue).

As far as I can remember you have always been limited to: model, make, lens and ISO (and some minor fields). An option to manually set a known metadata field name (for example: Exif.Photo.LensModel) and a specific trigger for that field would be a nice to have. No intricate existence check or fault detection needed: the responsibility lies entirely with the user if this "open" field and trigger option is used).

Just a thought.

[blueglyph]

When I see the list of models just in the case of Nikon, in the exiv2 project... I wonder if there's a simple exiv2 command to print out those columns for a new lens.

@kmilos Wouldn't it make sense to drop this issue here and that I make a request for a new Voigtländer lens in the exiv2 project, with a sample file? Could that translate the tags we discussed as desired and with more accuracy?