[Feature] Smart merge of duplicated assets #10846
Replies: 5 comments 4 replies
-
We plan to introduce an image quality assessment model that can detect the image with the least compression artifacts, least blurriness, etc. It can be reused in other cases as well, such as when generating person thumbnails and allowing users to search explicitly for low quality images.
This is possible, but EXIF data is unfortunately not always reliable; there are cases where the date will be set to 1970, 1900, some other incorrect date because the camera was never configured, etc.
Adding kept assets to the albums of duplicate assets is planned. I imagine merging EXIF metadata would be non-trivial since metadata for a duplicate may not be appropriate for the original, and again there may be mistakes in the metadata itself. If we handled it automatically, it would inevitably lead to mistakes that are hard to notice until later. If added, I think this would be something you configure rather than us opaquely deciding which metadata field to use from which asset. |
Beta Was this translation helpful? Give feedback.
-
From my experience here as a user of only 1 week, I would have liked a tickbox to prioritise a smaller HEIC over a larger JPG file in the dupe checker. We are a family of iphones and in the old days we connected them to the computer, and imported the HEIC files and they converted to JPG's as part of the backup process we used. Now in immich we have a JPG and a HEIC of many thousands of photos after an immich-go import of the filesystem on the PC and the icloudPD backups from the phones. After this latest version of immich, I let it mass delete the dupes, and hence it mass deleted all the HEIC's (as they are smaller that the corresponding jpg). Then I run immich-go again to select my icloudPD folder for the iphones and a large number of the HEIC's are back again - which would not have happened if the dupe detector prioritised HEIC. HOWEVER, I did this process again (run mass dedupe, then added icloudPD again) and this time no new dupes were created....... not sure what happened there - may have been an immich-go issue. I have repeated this import over and over from each set of icloudPD folders (from each iphone), and no more dupes have been added. I would like my workflow to be thus : I believe for my planned workflow, immich will work fine for me if there was a setting in the dupe detector to override HEIC over JPG. Once thats there, I will re-setup all the immich config from scratch and re-import all sources, then dedupe again. Hence keeping the HEIC duplicate instead of the JPG. |
Beta Was this translation helpful? Give feedback.
-
I really appreciate the dedupe feature! However, I would find it even more useful if there was an option to hide duplicates instead of deleting them. This would allow users to review duplicates later without permanently losing any data. Is this something that could be considered for future updates? Thanks for your hard work on this project! |
Beta Was this translation helpful? Give feedback.
-
To add, I think there is also the use case where we have pictures taken in sequence - similar to live photos, but manually taking pictures in sequence, for example of a group of people to then pick the best (i.e. one where no one has closed eyes, pulling faces etc).
|
Beta Was this translation helpful? Give feedback.
-
I think that merging metadata will be a really good idea |
Beta Was this translation helpful? Give feedback.
-
I have searched the existing feature requests to make sure this is not a duplicate request.
The feature
Currently, by default, duplicated assets are deduplicated by leaving the biggest file.
There is a big room for improvement, especially when using bulk deduplicate function, because right now there is no way to determine, whether bulk deduplicate is leaving the right file as original for each group.
In general, the goal of smart merge is to preserve all the best from duplicate group - highest quality content, maximum metadata, oldest dates.
Platform
Beta Was this translation helpful? Give feedback.
All reactions