Face detection - diffences between File Window & Viewer?

[nacho02]

Hello,

appreciate the great tutorial on the subject of face detection and recognition. :-)

Just had a couple of instances where I saw a behaviour which I could not quite understand.

I have an image which was not picked up as having faces when selecting files in a folder and running the Detect Faces and create annotations command in the file window. I'm ok with that.

But, if I open the file in the viewer and run the detection from there, it does detect faces. oh?

I'm suspecting this may have to do with the fact that the file in the file window is being displayed as being a CR2 File with mesurements of 1.600x1200 pix, but in reality the file is 4000x3000 pix, but there is an issue with the embedded preview being not full size. MY assumption is that the viewer uses the full res. file and detects the faces..

IS this the case or is there a difference in running the detection and annotation from both places?

Thanks in advance,

Ignacio

[Mario]

1. The Viewer always uses the cache image produced by IMatch. The default minimum size is 1920 pixel. If the screen rez is > 2000 pixel, the default minimum size is 2000 pixel. Check Edit > Preferences > Cache for the settings you have configured.

2. The Viewer always uses the "Optimize for small images" settings as documented. When you run a face detection manually, you need to use the same setting, else there will be a difference for faces smaller than ~ 100 pixel.

[nacho02]

Hi Mario,

thx. for the reply.

My Cache is set to 1920px. So the viewar is using a larger file than the CR2 file.

I'll see if I can reproduce this with other pics and get back to you.

Best regards,

[nacho02]

I can reproduce this, even with non CR2 files.

I have a JPG IMage 1551x1065px. I run the Detect Faces and create annotations command in the file window and I get one face detected and recognized. See attached file named "from the file window" (section of the file only showing the results).

I use a copy of the same image and run the same command from the Viewer. Result --> Image attached "From the viewer". I get all four faces detected and one recognized (which I'm ok with).

Am I thinking along the wrong lines or my expectation would be that both results are identical?

Thx,
Ignacio

[Mario]

When you run FD in the file window, do you use the option "Optimize for small faces"?
Do you look at the image in the Viewer at 100% or zoomed in?

[nacho02]

When you run FD in the file window, do you use the option "Optimize for small faces"?

No. Default value. Normal.

Do you look at the image in the Viewer at 100% or zoomed in?

Zoom to fit. But at that image size, it is 100%.

[Mario]

No. Default value. Normal.

The Viewer always uses the "optimize for small faces" option. Which explains the difference.

[nacho02]

Thank you. Good to know 

[Mario]

This is explained in the People help topic.

[nacho02]

This is explained in the People help topic.

my apologies if I led you to analyse something which was not needed, but if you mean this here:

https://www.photools.com/help/imatch/#people.htm#a_options?dl=h-101

I could not read anywhere that there is a difference betw. viewer and file window, and that "The Viewer always uses the "optimize for small faces" option."

Appreciate the clarity.

Thx.

[Mario]

In Special Cases I wrote:

You can also run face detection while in the Viewer. This always uses the 'Optimize for small faces' mode.

It's also explicitly mentioned in a highlight box in Manual Face Annotations:

When you run a face detection in the Viewer, it always uses an image size optimized for detecting small faces.

I know that the People help topic is rather long and detailed. But the "People" feature in IMatch covers a lot of functionality. This is a very rich set of features.

In general: If the faces in the images you process are less than ~ 100 square pixels (which is rare), use the optimize for small faces mode.