FAQ

How can I support this project?

Please read the CONTRIBUTING guidlines.

How to synchronize or upload my photos from mobile device?

HomeGallery does not support file synchronization from mobile devices and leave this topic to other great solutions.

We recommend to use SyncThing to synchronize your photos from your devices to a single NAS.

How to backup my photos?

HomeGallery does not support backups for your pictures and leave this topic to other great solutions.

We recommend to use BorgBackup or Restic for your self hosted backups. They support deduplicated, compressed and encrypted backups.

How to backup my gallery?

At bare minimum you should regulary backup your events.db file in your configuration folder. events.db contains your manual tag events. All other files can be recreated.

You should backup your configuration file, configuration folder and your storage folder. The configuration folder holds the file index files, the database, the events and the logs. The storage folder holds all preview files.

What data is requested from public services?

The goal of HomeGallery is to use as less public serivces as possible due sensitive private image data. It tries to use service which can be deployed local. However the setup requires technical knowlege and technical maintenance. Following services are called:

For geo reverse lookups (resolve geo coordinates to addess), HomeGallery queries the Nominatim Service from OpenStreetMap. Only geo coordinates are transmitted.

For reverse image lookups (similar image search), object detection and face recogintion, HomeGallery uses the its own public API at api.home-gallery.org. This public API supports low powered devices to run HomeGallery such as the SoC Raspberry PI. All low resolution preview images are send to this public API by default. No images or privacy data are kept from this API.

The API can be configured and ran also locally or as Docker container. See API Server section for details.

How many data is stored in the storage directory?

The media storage holds all extracted meta data, preview images and videos. See data structure for details. Videos are transcoded to a web friendly 702p video and preview images are resize to full HD (1920x1080) and lower.

Therefore, the memory consumption on the storage directory depends on the amount of images and videos. A rule of thumb would be 20% of your original space.

How long does the initial setup take?

It depends on the amount of images and videos and on your host CPU power. On the initial run all preview images and videos needs to be calculated. See design desisions for precalculations.

For images you have think in hours or days. For videos in days or weeks.

Use following exclude file to process only image files (JPG and PNG):

# Include only jpg and png files. Use this exclude file for quick start since
# video conversion takes time

# Ignore all files
*.*

# Allow directories with dots
!*.*/
# Allow jpg and png files
!*.jpg
!*.JPG
!*.png
!*.PNG

Due the internal structure already processed previews are not recalculated later and this needs to be done only once. Even if the image files are renamed or moved to other directories and their file content does not change.

Why is a progress indicator for all files is missing?

TL;DR the current non-progress state is on purpose. The recommended import is the initial (incemental) import via ./gallery run import --initial. As of v1.10.0 the source directories are automatically watched and imported by the server by default.

Looong version: Progress information on long taking processes like importing a gallery data is awesome. The assumption is that a user imports 50k - 100k media files (0,5 TB - 1 TB data) with some videos and wants to see as early as possible some pictures on the gallery.

However due the architecture of HomeGallery the files need to be indexed and the file checksums need to be calculated which are used as unique file ids. Afterwards the preview files and web compatible video previews need to be calculated. As final step the database is build. So there is no intermediate step to shortcut to present pictures in the browser.

If the whole media archive is processed at once it is time consuming for checksums, image and video previews. As mentioned, the image previews will take days, the video previews will take weeks depending on your amount of pictures, videos and host machine.

Since the assumed standard user wants to see early results, the recommended import of an archive should be incremental (or by chunks) via ./gallery run import --initial. Than the files are indexed and processed in batches and chunks and the database is build successive. The gallery will reload automatically when the database changes. So the user sees some pictures early while the whole archive import is in progress for weeks.

As of v1.10.0 the import of source directories is done in 3 batch steps:

• Import all files smaller than 20 MB which targets mainly images

• Import all files smaller than 200 MB which targets short videos

• Import other files for longer videos which needs time for transconding

Each batch is done in chunks (some file amounts). E.g. on the first run only 10 files smaller than 20 MB are imported. Than 20 files will be imported than a bit more. The chunks increase depending on the file amount in the file index.

The batch order and the chunk sizes ensure that a user get reasonable feedback of the process while fitting the internal data structure and building blocks of file index, extractor and database builder.

The internals of this progress is done via the indexer and the --add-limits parameter listed by ./gallery index -h (see also the source comments). This uses the gallery characteristic that the gallery knows only the files, which are indexed or known by the file index. So the indexer stops after adding some new files (afterwards these new files are processed, the database is updated, new files are indexed and so on and so forth).

How does the image similarity work?

The image similarity or reverse image search works through a pretrained image net of machine learning. It uses pretrained Google’s mobilenet to extract a similarity vector which is than compared agains other images.

In detail is uses the embeddings layer for the similarity vector and optimizes (or reduces the size) for the browser. The embeddings layer is retrieved via the API server as extractor step. When the database is build, the raw vector is reduced in data - in number of data points and in precision.

The exact algorithm where discovered in an investigation/analysis phase. As result it is more important to have data points than the precision. Therefore, the initial vector of 1024 floats is reduces to a third of 341 numbers with a 2 bit precision. These bits are packed to bytes and encoded in base64 to a JSON string with 116 chars like:

EGUUohiokFBoKUWwUhsFdRjUAEQQARElJZQmDlVCK6lgwsTGm2SRUYYEQGgAVRCCVEoQJFABVbJRSEIoQ1rpHDQUCIAWnVQRTlVSFIlDZQUEhLiFgAA=

With this procedure it is possible to reduce 1.3 GB raw vector data of 100,000 images to about 11 MB with some quality reduction but better browser support.

The comparison is done by the cosine similarity from a seed similarity vector against all other images. In modern browsers - even on mobile devices the similarity comparison takes less than 500ms for about 100,000 images.

Exact limits, factors or scaling should be extracted from the source.

I want to start all over, how do I reset the gallery?

There are cases where you want to start all over. Some cases are:

• You evaluated the gallery and want to import all your data

• You use your private API server via docker compose

• You migrate the gallery to another host

It is recommended not to delete everything and starting from a blank page. Precalculation of videos and images are expensive.

The best way for a reset is to keep following files and directories

• gallery.config.yml

• events.db (if any)

• The storage directory

This keeps your configuration of the source directories, your manual tags in events.db and the precalulated preview images and videos from the storate. The hard manual work of tags and the CPU intense previews can be reused again.

Than you can just delete the file indices (all *.idx files) and start the server. The server will automatically start to re-import your media.

Do I need to delete the storage directory?

Usually you do not need to delete any file in the storage directory. In the strorage directory all previews and meta data are kept. During processing existing previews and meta data are recognized and skiped.

You should only delete this storage if you migrate the storage or you like to delete the HomeGallery files (e.g. switching to another gallery solution).

Entry files cache is missing

If your stumble across a Entry files cache is missing error like:

extract:meta Entry files cache for
4d6c610:Pictures:Themen\Wallpaper\marriage toni\RIMG0056.JPG is missing.
Read files and meta from storage +4ms

The reason for this message is that the extractor most probably did not processed this file. This might happen due an incomplete or broken full media processing.

Are Raw Images supported?

If a RAW image file (or unsupported image format) has a JPG sidecar file, HomeGallery will use this JPG file for preview. E.g. IMG_1234.cr3 has a JPG sidecar file IMG_1234.jpg or IMG_1234.cr3.jpg than preview images are generated from IMG_1234.jpg (or IMG_1234.cr3.jpg).

If no JPG sidecar is available, HomeGallery can extract embedded preview images from the RAW image file through exiftool if available. These previews have mostly a lower resolution than the original image.

Note

It is assumed that disk space is not an issue for the usecase of HomeGallery. It is recommended to generate JPG sidecars by your own to have good image preview for your raw images.