beenull/photoprism
1
0
Fork 0
Code Issues Pull requests Packages Projects Releases Wiki Activity

Commands: Update face options formatting in show_options.go

Browse source 
Signed-off-by: Michael Mayer <michael@photoprism.app>
This commit is contained in:
Michael Mayer 2022-08-10 17:24:32 +02:00
parent 104d0a80d7
commit 8aef5670f6
1 changed files with 1 additions and 8 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

9
internal/commands/show_options.go
View file

@ -20,14 +20,7 @@ var ShowOptionsCommand = cli.Command{
}
var faceOptionsInfo = `!!! info ""
To [recognize faces](https://docs.photoprism.app/user-guide/organize/people/), PhotoPrism first extracts crops from your images using a
[library](https://github.com/esimov/pigo) based on [pixel intensity comparisons](https://dl.photoprism.app/pdf/20140820-Pixel_Intensity_Comparisons.pdf).
These are then fed into TensorFlow to compute [512-dimensional vectors](https://dl.photoprism.app/pdf/20150101_FaceNet.pdf)
for characterization. In the final step, the [DBSCAN algorithm](https://en.wikipedia.org/wiki/DBSCAN)
attempts to cluster these so-called face embeddings, so they can be matched to persons with just a few clicks.
A reasonable range for the similarity distance between face embeddings is between 0.60 and 0.70, with a higher
value being more aggressive and leading to larger clusters with more false positives.
To cluster a smaller number of faces, you can reduce the core to 3 or 2 similar faces.
To [recognize faces](https://docs.photoprism.app/user-guide/organize/people/), PhotoPrism first extracts crops from your images using a [library](https://github.com/esimov/pigo) based on [pixel intensity comparisons](https://dl.photoprism.app/pdf/20140820-Pixel_Intensity_Comparisons.pdf). These are then fed into TensorFlow to compute [512-dimensional vectors](https://dl.photoprism.app/pdf/20150101_FaceNet.pdf) for characterization. In the final step, the [DBSCAN algorithm](https://en.wikipedia.org/wiki/DBSCAN) attempts to cluster these so-called face embeddings, so they can be matched to persons with just a few clicks. A reasonable range for the similarity distance between face embeddings is between 0.60 and 0.70, with a higher value being more aggressive and leading to larger clusters with more false positives. To cluster a smaller number of faces, you can reduce the core to 3 or 2 similar faces.
We recommend that only advanced users change these parameters:`