Metadata in files that are sent outside?

[sybersitizen]

I'm wondering about a couple of details regarding files that are sent to outside entities for autotagging:

What are the pixel dimensions?

Are any metadata stripped before sending? More specifically, if the source file already has 'personally' entered metadata like IPTC entries, or keywords, or GPS coordinates, are those things stripped before sending?

The answers might allow some of us to rest more easily about privacy concerns with those services.

[Mario]

If you use the IMatch AI, no data is sent anywhere.

If you use one of the commercial services, IMatch sends the thumbnail it maintains for each file to the service.
The thumbnail contains no metadata.

[sybersitizen]

Thanks. And what are the pixel dimensions of the thumbnails? I haven't seen a way to access them directly.

[thrinn]

You configure the thumbnail size when creating a new database. So it should be this value (not sure if one dimension is less than the value configured here if the picture is not square).

[Mario]

The default thumbnail size in IMatch is 300 pixels.

[sybersitizen]

Thanks again, that's interesting. Yes, mine have been set to the default.

I wonder if autotagging from outside services would be more precise with slightly larger thumbnails. I only want to send the minimum resolution that allows good analysis, so maybe 300 is fine.

[Mario]

I've had a quick look at Google seems to have increased the required image size.
They now recommend 640 x 480  for landmark detection and label detection (keywords).
These  sizes were much smaller (around 300 pixel) the last time I've looked.

I might consider that when I update the Google code the next time, fetching and uploading larger thumbnails from the database.
So far the results with 300 pixels were very good for Google, also in comparison to other services.
Not sure if sending 4 times as much data will improve this much.

You can make a test easily:

1. Change the thumbnail size to 800 pixels under Edit > Preferences > Database.

2. Select a folder with some representative images

3. Select the files in the folder and press Shift+Ctrl+F5 > Force Rescan
Note: Write-back before if the files are pending.

4. Select the same images and run AutoTagger.
The Google connector does not manipulate the thumbnails and send them in the new size

5. Compare with your previous Google results for these images.

Let me know what you find out.

Don't forget to reduce the thumbnail size afterwards. 600 pixel thumbnails are usually not needed and create a much larger database.