There are times the software does not recognize a face. There should be a tool that allows you to select the face.
Yes, I have noticed that too that sometimes it does not recognize a face. I have had it happen a few times even when the face is large and looking right at the camera. I wish I could force Photo AI to detect that face when it doesn’t automatically.
Recover Face are designed for small & low resolution face only.
If the face is large & high resolution, “Recover Face” will only make the face blur, that is why TPAI automatically disable “Recover Face” when the face is large.
Actually, if you reread you will note that I did not use the words “high resolution.” It could have benefitted from Recover Faces as did the faces that were detected right next to it that were almost as big and were detected.
Thank you for the clarification.
In such case, I recommend you send the photo to developer, so that they can modify the face detection model. You can find the dropbox link in TPAI release post.
If the face detection model fail, manual selection won’t help much. If you force the software to put eye, nose, mouth in selected area, it will end up puting them in wrong place.
Yes - I have added my vote. I have several instances where faces are missed - usually because they are too small in the full image. When the image is cropped and reprocessed these missed faces are found as they are now larger wrt the whole image.
New to the program - Any updates on this feature?
I will say that the auto detect feature is incredible and works a majority of the time. The only time I have issues with it is on the following circumstances:
- older low quality digital photos (early 2000s)
- blurry photos
- photos that contain lots of patterns, especially when those patterns reflect natural objects. (such as a floral/victorian style couch)
Even under these conditions, the auto detect for the subject still generally works well. Plus, you can manually modify the subject.
Speculation on why this isnt a feature, based on my limited experience: The AI model is probably dependent on the auto-detect for faces. Meaning that for the AI ehancements to work, its contingent on the algorithm it uses to detect/optimize faces. If you were to do a custom face, I could see it throwing a wrench in how the face preservation system works. I’m sure its a concept they are working towards… Its probably just early in development and given the scope of the feature, I could see is delivering inconsistent performance. This is just speculation.
I will do some research and see if there are pre-optimization methods or programs that could be applied before using Topaz, to potentially help the problem. If I find something I will leave another reply with the info.
