Duplicates from re-upload after metadata changes not detected

[Terrance]

I have searched the existing issues, both open and closed, to make sure this is not a duplicate report.

• Yes

The bug

My workflow is to take photos from my phone, and later tag them via digiKam (with bidirectional sync of my camera folder). This means Immich will upload the untagged image at creation time, and upload a duplicate tagged version later.

As I understand it, the Duplicates utility is supposed to identify identical images, but requires ML (which I don't have configured) to do visual duplicates within a threshold. Currently none of these tagged duplicates are listed (despite a matching asset.thumbhash in the database), and the utility has only ever shown No duplicates were found.

The OS that Immich Server is running on

Arch Linux

Version of Immich Server

v2.1.0

Version of Immich Mobile App

v2.0.1 (F-Droid)

Platform with the issue

• Server
• Web
• Mobile

Device make and model

No response

Your docker-compose.yml content

n/a (installed via AUR)

Your .env content

# /etc/immich.conf
DB_HOSTNAME=localhost
DB_DATABASE_NAME=immich
REDIS_HOSTNAME=localhost
IMMICH_MEDIA_LOCATION=/mnt/data/media/own/immich
IMMICH_MACHINE_LEARNING_ENABLED=false
IMMICH_MACHINE_LEARNING_URL=http://127.0.0.1:3003
MACHINE_LEARNING_CACHE_FOLDER=/var/lib/immich/.cache

Reproduction steps

1. Upload an asset
2. Add/update an EXIF tag with digiKam
3. Upload it again

Relevant log output

Additional information

No response

[mertalev]

This sounds less like a bug and more like a feature request to make the duplicate utility work when machine learning is disabled.

[Terrance]

To confirm then, is there currently no duplicate detection at all when ML is disabled? The utility should probably be hidden according to ML configuration if it's never going to flag anything.

[Terrance]

I ended up digging in to the code myself -- no, duplicate detection is completely skipped when ML isn't configured.

I've had a go at implementing this in #23234.