This definitely doesn't work for me. I deleted the faces cache file (meaning that all my face tags went!) imported a single folder with about 80 photos containing very good single portraits and the analysis didn't recognise a 'face' in hardly any of them!

Phil

Hello Phil,

Over the time since PhD was first released, the face tagging has certainly improved a lot. It does, currently, throw up some odd results but my experience is that it's the exception rather than the norm.

Those odd results include:

• failing to recognise some very obvious faces (as you've found)


• interpreting objects like pockets as faces

As part of my regular workflow with PhD face tagging is a rarity but, after your post, I ran some "tests" here with PhD6. I imported 160 photos, all containing faces (2-6 in each photo). PhD ran it auto face detection and picked up most of them. I assigned names, sometimes using those suggested by PhD. In this step, I only had to skip 1 photo where PhD thought somone's elbow was a face After that I went back through, manually tagging about 20 faces it had missed.

Agreed, imperfect, but not like your experience. Would it be possible to attach a group of (even lower res versions) the photos you tried to tag. I'm not suggesting you may have done something wrongly at all. I'm just intrigued about how you got such disappointing results.

PIX PIX YouTube channel

Thanks Pix

This is really weird. To test the tag editor i chose a folder of holiday snaps containing typical family 'portraits' and various scenic shots etc. 87 photos in total.

I exported a selection of the photos, family faces only and created a new folder. I ran the face tag editor and it picked up all the faces perfectly.

I then did the same thing on the original folder containing the 87 photos and it picked up hardly any although some faces (e.g. in the crowd) from photos that I did test!

To overcome the face tagging issue I have started to create ordinary tags with the names of those in the photos. However do you know whether these tags are automatically written to the exif/metadata? I think not because when I open the same photos in Picasa or Windows live they don't appear. Yet if I do it the reverse way around and remove/then reimport the file intoPD6 then they do?

Do you have to export to write the new tags/keywords into the metadata? If you over-write (to avoid duplicate copies) is a consequence of this the degrade (compression) of JPG's even if only slightly?

I obtained PD6 for 'free' when upgrading Power Director 13 and wasn't really fussed - thought it would just be another photo manager. However I've generally been impressed with the user interface and some of its functionality. Just seem to be some funny quirks.



Thanks

Quote:

Do you have to export to write the new tags/keywords into the metadata? If you over-write (to avoid duplicate copies) is a consequence of this the degrade (compression) of JPG's even if only slightly?

Having done a bit of a test I think the answers are yes and no in that order!

Phil

Hello again Phil,

I'll do some testing on your idea that the image content of imported photos (i.e. people or mix of people & scenery).

You've answered some of your own questions by exploring. That's the best way to get an answer that will stick!

When you export photos from PhotoDirector there's a set of export options worth looking at. These set various parameters. You may have left these at their default settings.



When I export face tagged photos & include EXIF data, face tags are viewable in other software, including Picasa. Just select the photo, right click Properties & scroll down...



JPEG files are infamous for quality loss after editing. You can minimise this in your export settings. This (link) may be of interest.

PIX PIX YouTube channel

Phil,

I've not tested thoroughly enough, yet, to draw any conclusions but here are the steps & results of my initial test.

New project with people photos only [80], with 2 - 6 people in each photo - approximately 270 faces altogether. PhD tagged most of the faces, but missed ~20, which I manually corrected. In that project, 1 object (a hand) was incorrectly recognised as a face.

In a separate project, including the same 80 photos (above) and 65 other photos with NO faces, PhD missed ~25 faces & incorrectly identified 3 objects as faces - but not the hand it had identified previously!

I don't count these results as definitive in any way, but they possibly support your earlier statement.

PIX PIX YouTube channel

Quote: Phil,

I've not tested thoroughly enough, yet, to draw any conclusions but here are the steps & results of my initial test.

PIX

Thanks for your interest Pix. It does seem that there are some anomolies with the face recognition and that it may not be completely thorough. This may be understandable to some effect but I wouldn't expect it to miss obvious clear faces like it seems to have.



Phil