P: Face Aware Liquify doesn't find faces in some cases

[Happy Cranker]

With the new face detection, which is really great and fast tool to use, I have a lot of picture with the model reclining and the face is not detected automatically, is there a way to force the detection? In the end I rotated the picture 90 deg and worked on the face, with no problem.

[Kedar Bhat]

Photoshop liquify filter failed to detect an obvious face from the photo attached

 

[Meredith Payne Stotzner]

In the current implementation, our face-detection can recognize faces to about 30o rotation.  The suggested workaround is to temporarily rotate the image. Please let us know if the failure is occurring on faces with less than 30o rotation.

Additionally, the face-detection technology is not always able to recognize faces partially obscured by hats or head wear, extreme lighting, or face-paint etc. 

We are interested in seeing the images where face-detection isn't successful so we can make improvements to the algorithm in the future.

Thank you,
Meredith
Photoshop Team

[Meredith Payne Stotzner]

In the image Kedar Baht provided, I was able to determine the man's head wear is the reason the face is not detected. I did some testing where I merged the provided image with an image of a face without head wear, masking out only the head wear to reveal a forehead. The man's face could then be detected and manipulated, then masked back into the original image. (Note: the face-detection also had difficulty determining the center of the very squinted eye.)

[336home]

The attached face is not recognized. Would it be possible to enable manual facial control points when auto face detection doesn't recognize a face?

[Kedar Bhat]

Seems head gear & beard is not recognizable

[336home]

That's why it would be good to be able to manually identify control points when auto fails.

[Happy Cranker]

Also would it be possible using the red eye detection technology, be able to move the pupils left and right, rather than with liquefy only being able to move the whole eye? Do not get many models, who are cross eyed, but something like this would really help !

[mary@Lumos]

Meredith, I'm sending you an image where Face Aware Liquify doesn't detect the face.  

[mary@Lumos]

Here's the original image, which doesn't work with Face Aware Liquify either.

[DigitographyLive]

Here is an image that does not work

[Kedar Bhat]

Yes did not work as it is. Removed cap and it worked

[DigitographyLive]

Yes, from reading other comments, I masked out the cap and it worked fine.  Thank you

[FeedbackCommunityMember]

this image is not being detected

[ssprengel]

I don't have experience with face-liquefy, and this is not ideal, but this last woman's face is detected if you rotate it to make the eyes level...draw between the center of the eyes with the ruler and select Image Rotate Arbitrary (14-some degrees).

[FeedbackCommunityMember]

Face aware could not find the face in this photo.

[FeedbackCommunityMember]

YES!

[FeedbackCommunityMember]

it wont work on mine and shes facing forward 

[336home]

I wish Adobe would just enable manual identification of faces when auto-detection doesn’t work. It’s been about a year and there’s no progress on a fix 😞

[FeedbackCommunityMember]

> I wish Adobe would just enable manual identification of faces when auto-detection doesn’t work. >

I agree. On1 has a face detection algorithm that does allow us to manually select the face and reshape the paths for eyes and mouth. But On1 is editing for tone, color, and smoothing, which is much easier to mask for. 

I'm guessing that there's more to it when the entire face structure gets modified with one slider or another — so if the face is in strong profile or strongly tilted in any direction, how does the slider affect the underlying structure so what would normally make a smile from a mouth head on can do that with a mouth on a slant? Or half a mouth? I think it's possible for them to figure it out, but maybe it's taking this long because they don't have  one person assigned to it full time, and there's a lot of if-then stuff that has to go on under the hood to tell the sliders what to operate on and by how much.

They could have us use something like the On1 method to locate the edges of the features (don't forget it includes nose, jawline, distances of features from other features, and their shapes), but programming the sliders does sound like it could be very complicated.

But for more simple issues like the one right below, where the program can't tell eyes from glasses, or the downward tilt, or maybe the beard, (and I know it has problems with hats, too), I'd be thrilled with a hybrid between On1 and the auto recognition we have now, even if they can't solve all the problems.

I do hope they're working on it, anyway. It's a great convenience not to have to use the other tools to make the alterations. I'm not expecting much anytime soon, though, since I very much doubt they assign one such project per person, but I'm always hopeful.<G>

[Hannah Nicollet]

Please download the new release of Photoshop that became available this morning. I tested your image and face-aware detects it now.

[Hannah Nicollet]

This has been addressed in the latest release of Photoshop, which became available this morning. I tested one of the images in this set and face-aware had no trouble detecting the face in the image. Please try it out!

Thanks,
Hannah Nicollet

[336home]

Unfortunately you didn't test it on the face I posted 😞
Again, would it be possible to enable manual facial control points when auto face detection doesn't recognize a face?

[Hannah Nicollet]

Hi Gordo,

I tested your image and face aware had no trouble with it that I could see. What problem were you having?

Manual face selection isn't available at this time.

Thank you,
Hannah Nicollet

[336home]

The problem was that it didn't recognize a face (I'm using: PShop CC 2017.1.1 Release 20170425.r.252 x64 updated this morning.

As you can see:

[Hannah Nicollet]

I tested it on Mac and it worked but it looks like you're on Windows? I'll test it over there.