Borg backup of Nextcloud Loose Previous Tagged Faces
还没有人认领这个 Issue。
评估
- 难度
- 5/5
- 预计耗时
- 一周以上
- 新手友好度
- 15/100
调研方向
The payload names no source files, tests or entry points, so the code location is not given. Start by working out how Recognize stores face clusters and person names in the Postgres database, and how its clustering job reassigns faces when it re-runs after a restore. Done would mean either a documented, tested export and restore path for those assignments, or a clear maintainer decision that the behaviour is intended. The reporter's request is a design question, so a first-timer should confirm the scope with maintainers before writing code.
由索引模型根据 Issue 内容生成。
描述
Which version of recognize are you using?
13.1.0
Enabled Modes
Face recognition
TensorFlow mode
WASM mode
Downstream App
Memories App
Which Nextcloud version do you have installed?
35.0.1
Which Operating system do you have installed?
Windows 11 Pro + Docker Desktop 4.95.0 / 29.8.2 on WSL 3.0.3.0
Which database are you running Nextcloud on?
Postgress 18.6
Which Docker container are you using to run Nextcloud? (if applicable)
AIO
How much RAM does your server have?
64gb
What processor Architecture does your CPU have?
AMD Ryzen 9 5950X - x86-64 / AMD64
Describe the Bug
After restoring Nextcloud from a recent AIO Borg backup, the photos themselves were intact, but much of the face-recognition work was not preserved correctly. Recognize had previously grouped faces, and I had spent considerable time manually identifying and correcting those groups. After the restore, some person names survived, but many were missing, fragmented, or incorrectly reassigned.
Because the Borg backup was current, the likely issue is not stale data. The more probable explanation is that Recognize restored its database state and then re-ran face processing or clustering, which altered some of the manually curated identity assignments. In other words, the physical photos were protected, but the human-added face identification and clustering state was not reliably preserved through restore and reprocessing.
That is the gap I am trying to solve: how to back up or export the Recognize face/person metadata independently so that a future restore does not require redoing hours of manual tagging.
Expected Behavior
After restoring from a current AIO Borg backup, I expected my existing Recognize face clusters, person names, and manual corrections to remain intact. The photos themselves were restored correctly, but many face assignments were missing, fragmented, or incorrect afterward, while some survived. Since Recognize is supposed to preserve existing cluster assignments and manual edits during subsequent clustering, I would not expect a restore/reprocessing cycle to materially alter previously curated identities.
To Reproduce
- Use Nextcloud AIO with Recognize enabled and allow face detection/clustering to complete.
- Manually name several face clusters and correct some incorrect face assignments.
- Create a current AIO Borg backup after those manual changes have been made.
- Restore Nextcloud from that Borg backup.
- Start Nextcloud normally and allow Recognize/background jobs to resume processing.
- Open Memories/Recognize People view and compare the restored face groups with the pre-backup state.
Observed result: some person names and face assignments are preserved, but others are missing, split into new clusters, or incorrectly reassigned.
Expected result: the restored face clusters, person names, and manual corrections should remain consistent with the state captured in the backup.
Debug log
No response
- 主要语言
- PHP
- 星标
- 701
- 派生
- 70
- 平均合并
- 7 天 12 小时
- 30 天内合并 PR
- 3
环境准备
- 没有 Dockerfile 或 Docker Compose 文件
- 没有 Pull Request 模板
- 阅读贡献指南
从这里开始
- 先读完整个 Issue，再读项目的贡献指南。
- 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
- Fork 仓库，在一个分支上完成修改。
- 提交 Pull Request，并在描述里引用这个 Issue 编号。
nextcloud/recognize 的其他 Issue
-
难度 4/5 3-5 天 新手友好度 48/100
-
enhancement
难度 3/5 1-2 天 新手友好度 65/100
-
enhancement
难度 4/5 3-5 天 新手友好度 45/100
-
Recognize maintenance routine unnecessarily redownloads Tensorflow CPU and GPU models every time可能重新可做 @marcelklehr 于 205 天前认领，目前没有进行中的 PR。 未关闭bug
-
enhancement
难度 5/5 一周以上 新手友好度 38/100
查看 nextcloud/recognize 的全部 Issue
相似的 Issue
-
难度 2/5 1-3 小时 新手友好度 76/100
维护者通常 1 天内回复
-
[2.x] extension-manager: translation keys that changed meaning keep their old names可能已有人在做 关联的 PR 仍在进行中或已合并。 未关闭type/bug
难度 2/5 1-3 小时 新手友好度 84/100
维护者通常 1 天内回复
-
needs-triage Technical TODO
难度 2/5 1-3 小时 新手友好度 68/100
维护者通常 2 天内回复
-
难度 2/5 1-3 小时 新手友好度 72/100
MISP/misp-docker#478 ·
维护者通常 1 天内回复
-
难度 2/5 1-3 小时 新手友好度 84/100
glpi-project/glpi#26045 ·
维护者通常 1 天内回复