dacr / sotohp

Photos management

SOTOHP

A software to easily and quickly manage a huge amount of photos. Why ? Just because those past 23 years, I've taken more than 90000 photos. Once again this project has first started as just a "small" script.

Principles

• The photos referential is always your photos directories as you've organized them,
• Your photos directories are left unchanged, only read operations are done,
• A local LMDB database is used to store/cache/query, serialization is JSON based,
• Background jobs are run to update, to enrich, to analyze, to process your photos
• Dedicated Filesystem targets for pre-processed photos
• Photos are identified using an ULID identifier computed from the shoot timestamp
  • if shoot timestamp is missing or invalid, file last modified is used (at first seen)
  • SO photos stream is automatically ordered by this timestamp
• DeepJavaLibrary is intensively used for
  • photos classifications,
  • people faces detection,
  • objects detection,...

Current status

• no release available yet
• code is stable and usable for
  • visualization
  • publishing photo records to elasticsearch / opensearch
  • searching and statistics purposes

The visualisation with the graphical user interface is very fast :) It starts in 2 seconds even with my 92000 photo database.

Published information to elasticsearch or opensearch make it easy to build very interesting dashboards such as :

sotohp's Issues

Use ULID also for photo owners

Improve missing shooting date strategy

Use in priority :

1. photo shooting date timestamp in exif data
2. use timestamp coming from a photo with exif data in the same directy
3. extract timestamp from directory naming rules
4. use file created date

photos object instance segmentation

See : https://www.reasonfieldlab.com/post/instance-segmentation-algorithms-overview

Support HEIC image file

Support photo rotation in viewer

• automatic orientation for normalized and miniatures
• provide option to rotate the photo using bracket keys in the viewer

photos classification

• wire legacy model
• store found classes

publish photos to elasticsearch/opensearch

• reuse photo legacy file records
• publish to opensearch using legacy code ? Or using zio-elastic ?

photos analysis

• photo files analyzer
  • originals (file format issues, ...)
  • generated photos miniatures
  • normalized photos
• results
  • summary report
  • detailed report

photos faces clustering

• extract faces features
• faces clustering
• associate face clusters with people names

Support symbolic links for search roots

Induce missing GPS coordinates

Using such approach

• add a field induced or realGPS field
• add an algorithm such as
  • consecutives photos coming from the same camera
  • approximate missing coordinates between photos with coordinates in a limited timerange
  • time base average between two known points (on a line, and depending of timestamp)

Support streaming from database

• add streams store operations
• add photo (from database) stream

Simplify generated photo artifacts storage directories

• just one hierarchy
• just one env parameter
• just one config env
• store relative file in database for generated artifacts

Support multiple photo owners

• add owners collection
• separate collections ?
• migrate current storage

photos faces detection/extraction

• detect faces
• store bounding boxes
• count how many faces (and so people)
• face features extraction and storage

text in photos extraction

• with an OCR approach ?
• use found as keywords ?

photo objects detection

• call legacy model
• enhance with a richer model ?
• store detected objects names
• store detected objects bouding boxes

Minimalist Desktop User Interface

• tech candidates evaluation
  • doodle : https://github.com/creativescala/doodle
  • skija : https://github.com/HumbleUI/Skija
  • compose : https://www.jetbrains.com/lp/compose-multiplatform/
  • javafx
• photos mosaic
• photo viewer with quick previous/next
• simple photo search
• quick edit/update photo description