Last Updated or created 2022-05-29

A few years ago i wrote a photo manager .. again .. ( see post about my first previous photo manager )
It is a web gui to find photos in my huge photo archive.
I manually added 190k tags to 120k photos in 20+ years.

I thought wouldn’t it be nice if i can generate additional metadata using Machine Learning. A few years ago i did some testing and followed a podcast and free course about machine learning.

So today i started to implement a addition to my gui. Machine recognition tags!

It already kinda works.

Things to do :

• Make it a background job, my fileserver doesn’t run Tensorflow on a GPU, so it is slooow
• Embed in existing GUI and stats
• Design a editor to remove wrong tags

Below a part of ML images

Command to get a thumbnail sheet with only directory names:

montage -verbose -units PixelsPerInch -density 300 -tile 7x6 -label "%d" -font Arial -pointsize 6 -background "#FFFFFF" -fill "black" -define jpeg:size=253x154 -geometry 253x154+2+2 -auto-orient */*.JPG -title "ML Thumbs" thumbsheet.jpg

Maybe, i can use debug output like below.

['lakeside, lakeshore (score = 0.47934)', 'seashore, coast, seacoast, sea-coast (score = 0.11385)', 'sandbar, sand bar (score = 0.08822)', 'breakwater, groin, groyne, mole, bulwark, seawall, jetty (score = 0.06281)', 'valley, vale (score = 0.01790)', '']