Recovery face does not consider excluded areas

I realized that Face Recovery does not take into consideration areas that are excluded from subject.
As result recovery is applied on the entire face but sometimes, especially with mustache and hair, this does not produce a good result.


Topaz Photo AI [v1.2.8] on [Window]

1 Like

Hey there,

It looks like you may be working with film scans of old photographs here, and unfortunately, the results for these photos are going to be very hit or miss for these photographs.

Our apps have been all trained on photography purely from modern-day digital cameras. In essence, this means that when our AI models come across a photo from a film camera, oftentimes it doesn’t know what it’s looking at. This is due to film photography (especially scans of older prints) behaving much, much differently than modern digital cameras. Things like film grain, scratches or tears on prints, and artifacts from scanning, all are factors that our algorithms have not encountered before when they were trained on digital images.

It’s possible for our software to have occasional success when processing film images, but the photographs would need to be extremely low-noise and would need to be free of all other blemishes. I’m afraid that if you are not getting notable results with Sharpen AI, and you’ve tested out all of the parameters within the app, the film photo you are working with is likely unable to be improved further due to the aforementioned limitations of our software when it comes to film photography. Unfortunately, film photography was never a use-case that we intended our software to work with.

Hope this can help clarify things a bit, please let me know if you have any more questions I can answer.

Thanks for the explanation Preston.
I used your software on a 1900 photo scanned at 600dpi.
I must say that your AI models have always given me good results on these photos and also on this face I must say it works quite well.

However, the model does not recognise the moustache well (perhaps due to the type of learning based on digital photos). Since the face is calculated, wouldn’t it then be possible to apply it only in the selected areas (red colour)?

That was my original question

Regards
Moreno

Ahh, my apologies, I misunderstood your original question! Thanks for clarifying – that is a very interesting idea, but not something that is possible within Photo AI at the moment. When face recovery is enabled, it will apply to the entire subject’s face. The masking you have there will only apply to the sharpening module, I’m afraid!

I think we’re currently training more models on hair to improve reconstruction of facial hair in the future, this (combined with the historical photo factor as you noted) may be why the results on the moustaches aren’t coming out as clear as we want them to be!

Would it not be possible to give the user access to the mask used by the facial recovery so that he/she may modify it so that the original (subject sharpened) image is revieled in selected areas? The facial recover sometimes affects only parts of the hair which gives uneven result, an sometimes the face recovery affects shoulders an parts ob background. Regards HAL

I too have reported the same problem in the past.
Unfortunately, facial reconstruction almost always leaves half of the hair in low definition and with a boxy shape. It would be really useful to be able to define the mask so as to avoid this annoying inconvenience