Face Recognition.

[SmplyTryn]

Lightroom Version

10.2

Operating System

1. Windows 10

I am attempting to enrich my catalogue using the face recognition system. However my catalogue comes from several sources including digital images, slides and negatives that I have developed using a lightroom plug in. As you know the negatives are stored in the lightroom system in their original format and the positive comes from Lightroom processing. It seems that face recognition will therefore store both negative and positive representations of the same person as I have different types of images for the same individuals. I intend to export the whole catalogue into an external device for distribution. The negatives evidently will not be present in the final product. Will that foul up any external classification that is used? Will it even work?
Thanks
Terence

 

[SmplyTryn]

Addendum I am concerned that the function to seek similar faces is also not working as I found that after using the box to identify a face, then asking lightroom to find similar faces produced not response even though there were more shots of the same person in an adjoining frame.

 

[RobOK]

I’m not clear what you mean about positives and negatives, aren’t they all image files?
Maybe share a screenshot if you can...

 

[Linwood Ferguson]

If you use something like negative lab pro which does a tone curve inversion to give you positives, those cannot be processed (correctly anyway) by facial recognition. If you then create derived images though photoshop, where it creates a TIF (or other) and then edits that and brings it back as a separate master, that will be able to be processed but rather defeats the purpose. Creating virtual copies will work just like originals (i.e. will not work).

It's a downside of negative conversion by tone curve inversion.

 

[PhilBurton]

If you use something like negative lab pro which does a tone curve inversion to give you positives, those cannot be processed (correctly anyway) by facial recognition. If you then create derived images though photoshop, where it creates a TIF (or other) and then edits that and brings it back as a separate master, that will be able to be processed but rather defeats the purpose. Creating virtual copies will work just like originals (i.e. will not work).

It's a downside of negative conversion by tone curve inversion.

Ouch, ouch again. For someone like me, with probably 10 thousand negatives to scan, this is a major kick in the gut.

If you use NLP without Photoshop, will you still have this problem? Of course, for scanned images, dust removal is a big issue and that is done better in Photoshop than Lightroom.

Is there a workaround, short of manually tagging all faces in scanned negatives? An alternative to NLP?

 

[Linwood Ferguson]

If you are doing the dust removal in photoshop I'm not sure there is a need to keep the negative, or conversely do you need to face-recognize the negative?

I know for old time photographers saying "why keep the negative" is heresy, but in this case it's really just a digitization of the negative. You could just delete them and face recognize the positives that come back from photoshop.

The sad fact is despite a small but vocal group continually asking for negative support, Lightroom is a very unfriendly environment for in-situ conversion of negatives, the order of steps in develop is all wrong. NLP does a great job of working around it, but it's the real lack of underlying support that is at issue, and is why face recognition doesn't work in it.

 

[Linwood Ferguson]

I'm sorry, I may have misread your note: you said if you use NLP without photoshop. It is without photoshop you have the problem.

With photoshop, i.e. digitize, run NLp to produce a displayed positive, edit-in-photoshop, and take that back as a new image in lightroom -- this last image is a real positive and should work with facial recognition. It's the NLP'd image that will not (nor will a manually inverted tone curve, it's not NLP's fault).

 

[PhilBurton]

I'm sorry, I may have misread your note: you said if you use NLP without photoshop. It is without photoshop you have the problem.

With photoshop, i.e. digitize, run NLp to produce a displayed positive, edit-in-photoshop, and take that back as a new image in lightroom -- this last image is a real positive and should work with facial recognition. It's the NLP'd image that will not (nor will a manually inverted tone curve, it's not NLP's fault).

Ferguson,

Thanks for this quick reply. Now I can breathe again.

I agree with the central point of your prior post. If my workflow starts with NLP in Lightroom, and then dust/scratch removal in Photoshop, then the resulting positive TIFF file IS my digital negative, and there is no reason to keep the actual negative scan in Lightroom. It is probably wise to keep those original negatives scans as disk files. And this workflow could be a justification for a separate Lightroom catalog just to do the NLP and Photoshop export work.

 

[SmplyTryn]

I am sad that my observations were correct and that without converting the images to positives the facial recognition software will not work. I wonder whether just exporting to an external file and then reimporting will also work. That is a way that corrects the reversal of the controls. Exporting the image I think will export the positive as that is how you transmit the final image to someone else though it is a lot of extra work. You would also need to have finished your development of the image.

 

[SmplyTryn]

A way of handling the problem without too much difficulty would exist if it would be possible to export a large folder of photos to an external folder, and then reimport it without having any dislocation of the organisation of these files. I wonder whether the lightroom catalogue would enable that?
It would then avoid having to deal with photos on an individual basis which would disrupt the work flow and would enable the processing of the whole catalogue as positives would not be damaged by the process. Does a light room expert have any information on that?
Thanks
Terence