Currently running Immich local network only (not yet public facing)
When browsing images i somethimes see faces tagged but the with a missing feature photo, this also causes those "people" to not show up in the people list.
Secondly i don't seem to have a "change feature photo" option (might be me being daft)
#No "change feature photo" option and some feature photo's are black causing people to be hidden
1 messages Β· Page 1 of 1 (latest)
You have to go to the people tab, click on a person, and then in the top right hit select feature photo
ah
thanks,
then there is still the issue of missing feature images and that causing, them not to show up in the list of people
That I am unsure of with no logs and such
if you tell me what logs you need and where to get em, ill try π
Maybe try and fix their featured photos by updating them and see if that resolves the issue.
Would need the microservice and server logs. https://immich.app/docs/FAQ/#how-can-i-see-immich-logs
manually setting a feature image, does fix the issue
however now finding all the people who he's "found" without a name is another thing.
Will try to get the logs π
there you go
Hmm. Nothing suspicious in there error wise. Not 100% sure what would have happened in this case. Can see if anyeone else has any insights
please do π
I think i know why this is happening
i just tried to change the feature photo and selected a video
and then the feature foto changed to a blank one
so could be that the face recognition is trying to get feature pictures from video's
Oh yes, I doubt that would work. Thumbnails are generated based off the first frame of a video, if there's no face in it that could cause issues. Let me ask team about it
causing this issue
Is it blank or broken?
Alright, asking the crew
thanks!
Seems like a bug, I would search github and make sure it's not a (open) duplicate and if not, open a issue with your setup, description and such π
will do
Thank you!
you too!
Same here. Seems like a bug and I really hope the broken faces can be fixed after the bugfix without having to re-recognize all faces π
This is what comes in my logs regarding this error:
[Nest] 310 - 06/12/2024, 12:55:59 PM ERROR [Microservices:JobService] Unable to run job handler (thumbnailGeneration/generate-person-thumbnail): Error: Input file contains unsupported image format
[Nest] 310 - 06/12/2024, 12:55:59 PM ERROR [Microservices:JobService] Error: Input file contains unsupported image format
at Sharp.toFile (/app/immich/server/node_modules/sharp/lib/output.js:89:19)
at MediaRepository.generateThumbnail (/app/immich/server/dist/repositories/media.repository.js:69:14)
at PersonService.handleGeneratePersonThumbnail (/app/immich/server/dist/services/person.service.js:429:36)
at process.processTicksAndRejections (node:internal/process/task_queues:95:5)
at async /app/immich/server/dist/services/job.service.js:148:36
at async Worker.processJob (/app/immich/server/node_modules/bullmq/dist/cjs/classes/worker.js:394:28)
at async Worker.retryIfFailed (/app/immich/server/node_modules/bullmq/dist/cjs/classes/worker.js:581:24)
[Nest] 310 - 06/12/2024, 12:55:59 PM ERROR [Microservices:JobService] Object:
{
"id": "fd9ccfcb-806f-493e-b16d-72cf5a58d4d4"
}
Is this a file you can share for troubleshooting?
it happens on all videos that were uploaded in the timesince last update (the one before yesterday) and only on faces which weren't already known
I just made a 1sek Video of the google images results of "person" and it happened again
it doesn't matter if the video came form my form or through web UI. The GoPro videos I uploaded are also affected
this is the log of the upload of this video
Up to you, I think it might be some artifact from previous versions that cause the issue
314 faces of all people in my life since I was born
isn't it possible to select all damaged faces from the database only
You can also spin up a test instance and try those new photos
Re recognize all faces of the new photos that have the issue is no problem
I haven't sorted anything there, as they are damaged anyway and I was waiting for a fix in the next version π
But "Input file contains unsupported image format" looks like an error in the thumbnail processing, not inside the database, doesn't it?
just checked my timeline again. it definetely began with version v1.105.1
where are the face-thumbnails stored? could it be a folder permission problem maybe?
They are stored in the thumb folder as well
Hi @polar karma I wonder if this is an edge cases for the change we made in cropping the thumbnail
I think the missing person thumbnails are related to people only identified in videos. I noticed some errors in my logs when I upgraded Immich and re-ran the missing thumbnails job. I get the "Input file contains unsupported image format" only for people that only appear in videos, but no photos.
Hmm do you have a video that you can share for reproducing the issue?
Sure, how can I send it to you?
You can zip it up and DM me
@short moth Did you change your video transcoding settings?
could be, but my settings look pretty "normal"
"ffmpeg": {
"crf": 25,
"threads": 0,
"preset": "fast",
"targetVideoCodec": "hevc",
"acceptedVideoCodecs": [
"hevc",
"h264"
],
"targetAudioCodec": "aac",
"acceptedAudioCodecs": [
"aac",
"mp3"
],
"targetResolution": "original",
"maxBitrate": "90M",
"bframes": -1,
"refs": 0,
"gopSize": 0,
"npl": 0,
"temporalAQ": false,
"cqMode": "auto",
"twoPass": false,
"preferredHwDevice": "auto",
"transcode": "required",
"tonemap": "hable",
"accel": "disabled",
"accelDecode": false
},
"machineLearning": {
"enabled": true,
"url": "http://127.0.0.1:3003",
"clip": {
"enabled": true,
"modelName": "ViT-B-32__openai"
},
"duplicateDetection": {
"enabled": true,
"maxDistance": 0.0155
},
"facialRecognition": {
"enabled": true,
"modelName": "buffalo_l",
"minScore": 0.7,
"maxDistance": 0.5,
"minFaces": 1
},
"classification": {
"enabled": true
}
},
I just updated to the hotfix but the problem still persists
It wasn't addressed in the hotfix because we don't know where is the issue yet. Do you mind reset your transcoding to default and try again?
For regenerate thumbnail for missing?
I always record a short video of the google image results to test it π
I'm using the All-in-one Image btw (but thinking about switching to the official release as it's compact now)
We probably still need to crop the preview image for videos