Closed darrepac closed 3 years ago
darrepac
commented
3 years ago Another comment: everything is smooth and works well on my lan. When I try from outside my lan, I cannot make it work (timeout) because my home internet connection is probably too slow...
xemle
commented
3 years ago I LIKE:
* the infinite scrolling that works very well and is reactive * the scan process that seems to work well comparing to other solution (librephotos typically that is damn slow on old machine) * video support works well
Thank you for your positive feedback. Glad you like it.
WHAT I MISS/DISLIKE (this is my opinion and to make things better, not criticism):
Do not hesitate to come up with improvement ideas. It helps to push the gallery into the right direction.
Keep in mind, that it is a spare time open source project and free to use. My time is limited and most features (or missing features) are a result of a prioritization of requirements.
My vision is stated at the landing page with some motivation and technical details on the documentation. The gallery solves already my biggest requirements: privacy/no cloud provider and exploring/discovering/stumbling/browsing/viewing the whole archive through the similarity/reverse image lookup.
Currently I am focus on polish the general usage through quite new binary distribution and the initial UX.
* even if I start the scan process from 0, I have the following (seems to be old photos that I deleted but are still present in the database??): 
It seems that the previews were not rendered. This might be caused by: a) missing full update with all images b) missing JPG/PNG of a RAW file c) errors on reading the imags
Do you have an idea in which error category it might fit?
* timeline missing. Ok you can choose images per year but would be nice to have some date on the side of the infinite scrolling (as are doing GPhoto, OneDrive, Apple, Plex etc) in order to 1/ jump more precisely to a date 2/know where you are
This is a nice feature request.
I have some ideas to improve the years selection by year -> month -> week. The same can be done for folders or geo data
* icons that are not always well readable:  
Yeah. It is hard to find a good solution to focus on images without to much disturbing of navigation.
One idea is to toggle the navigation icons on an of. So it is possible to show only the image or the image with more visible/disturbing icons.
* geolocation is in german ;) (at least english would be nice): 
This feature comes with the next release and is already on the master.
I expect in the next two weeks a new release with some initial UX improvements.
* no way to correct / improve face recognition neither a way to see al the recognized faces
Have you tried the i icon and clicked on the recognized faced marked by red boxes to see similar faces?
Face recognition is a hard topic. I am not to much in ML and I am depending on public open source solutions. The used library is faceAPI.js
The current state is more an proof of concept and I do not see much space for quailty improvements.
For me the face recognition is a good complementary feature to the similarity feature. Where the similarity feature fails, often the face recognition succeeds and vice versa.
* when I tried "edit", and tried to select 1 photo, it selects 92 photos.... so I did not try to edit anything, too risky ;) 
In edit mode you can multi select images and tag them. The selection stays for further edits because it is easier to keep the current selection and clear it than reselect the last selection. So you need to clean the selection if you like to start a new selection.
This behaviour is stated in the edit docs. Improvement suggestions are welcome.
* the geolocation feature is under used: a map with all photos in it (by cluster) would be nice
This is the result of prioritization triage.
As stated before I use the gallery mainly to navigate through my whole archive. Usually I do know where the images where taken and it is not so important to me.
IMHO this feature is absolutly awesome to show others a map with all the geo positions by not so much for an image owner.
* the object recognition is always funny, but here also we can't do anything with that. If I want to see all apple, how do I do? ;) 
The object recognition is very bad - I know. Sometimes it helps. Mosty not. So it is also not promoted a lot.
I have done this feature at the same time as the face regognition. If there are better solutions (based on tensorflowJS) I am happy have a look.
These are my cents of reply
Could you provide an ordered/prioritized list of your problems/requirements/use cases for your gallery usage to respect these in the next prioritization/feature triage? What should be done to solve you main pain points?
darrepac
commented
3 years ago * even if I start the scan process from 0, I have the following (seems to be old photos that I deleted but are still present in the database??): It seems that the previews were not rendered. This might be caused by: a) missing full update with all images b) missing JPG/PNG of a RAW file c) errors on reading the imags
Do you have an idea in which error category it might fit?
Well I checked one of those. It is a classic JPG files as all the others. Despite what I thought initially, the file is not a file that I deleted. In the terminal they are not much information. Is there a verbose mode?
* timeline missing. Ok you can choose images per year but would be nice to have some date on the side of the infinite scrolling (as are doing GPhoto, OneDrive, Apple, Plex etc) in order to 1/ jump more precisely to a date 2/know where you areThis is a nice feature request.
I have some ideas to improve the years selection by year -> month -> week. The same can be done for folders or geo data
Well I don't think that going through a sort of calendar view is not the way to go. You have to take profit from your infinite scrolling that works well, to be able to have all pictures in 1 page, with a timeline on the side.
The use case is simple (and the one I use the most):
I remember about something, it was about in July 2017.
Then you jump in July 2017, but may-be it was more June or August and so as you are in your 1 page view with all images, you can move up or down depending the need.
Generally speaking, when you scroll in your view you have no idea about the date. Look at this:
1: wherever you are in your "1 page infinite scrolling view", you know when
2-3: side timeline let you jump quickly into an approximate date
80% of the time, I use this
* icons that are not always well readable:  Yeah. It is hard to find a good solution to focus on images without to much disturbing of navigation.
One idea is to toggle the navigation icons on an of. So it is possible to show only the image or the image with more visible/disturbing icons.
Not a bad idea!
* geolocation is in german ;) (at least english would be nice): This feature comes with the next release and is already on the master.
I expect in the next two weeks a new release with some initial UX improvements.
ok
* no way to correct / improve face recognition neither a way to see al the recognized facesHave you tried the i icon and clicked on the recognized faced marked by red boxes to see similar faces?
yes I tried it
Face recognition is a hard topic. I am not to much in ML and I am depending on public open source solutions. The used library is faceAPI.js
The current state is more an proof of concept and I do not see much space for quailty improvements.
Well I was not clear. The idea is not to improve the library byt itself but the result:
Usually each library permit to "train" the model. Let's take an use case:
It detects my face on 70 photos. But on 5 of them, it is not me. There should be a way to say that those faces are not me but xxxx. On the other side, it could fail to detect me on some pictures, so we should have a way to say that thoses faces are me.
Then, there should be a page presenting all the faces detected. So we can select one and see all photos related to those faces.
Here also, Google Photos is a good reference:
10% of my time I am using this to find someone's photos.
* the geolocation feature is under used: a map with all photos in it (by cluster) would be niceThis is the result of prioritization triage.
As stated before I use the gallery mainly to navigate through my whole archive. Usually I do know where the images where taken and it is not so important to me.
IMHO this feature is absolutly awesome to show others a map with all the geo positions by not so much for an image owner.
Well in 10% of my photo search, I am looking for some photos in a specific places and do not remember precisely the date: here the map view is useful. I switch to map view and then can zoom in to the place I am looking for.
* the object recognition is always funny, but here also we can't do anything with that. If I want to see all apple, how do I do? ;) The object recognition is very bad - I know. Sometimes it helps. Mosty not. So it is also not promoted a lot.
I have done this feature at the same time as the face regognition. If there are better solutions (based on tensorflowJS) I am happy have a look.
Here my remark is not about the quality of the recognition library but more of the use. When you look at a photo and it says that there is bottle in it. As you can do with faces, we should be able to press on the bottle and see all pictures with bottle.
These are my cents of reply
Could you provide an ordered/prioritized list of your problems/requirements/use cases for your gallery usage to respect these in the next prioritization/feature triage? What should be done to solve you main pain points?
I think the number 1 feature is about the timeline/infinite scrolling. That's I think the people moving away from Google Photos are looking for (me included). The last 6 months I have tested many solutions and none arrive to do it. Yours seems not so far as you already solved the infinite scrolling part (but at a cost of a solid internet connection) Then face recognition / map view improvment (not in terms of library at this stage, but more in terms of use / ux) is also something people are looking. And to reach those people, I think you cannot avoid a docker solution, easier to deploy (with docker-compose / yml file).
xemle
commented
3 years ago Hi @darrepac would you mind to open separate issues for your requirements/wishes/desire to discuss them separately?
AFAIU these are the requested features ordered by priority and complexity
darrepac
commented
3 years ago Note: as you have seen in an other thread, I have started a new setup through docker.
But I end up with the same problem with some images:
xemle
commented
3 years ago Either the import step was not done/successful or the preview calculation failed as mentiond before.
Have you tried to test these files isolated by a separate source folder and separate gallery config to qualify that the extractor step works (or not)? A log of it would be helpful here.
You can also verify that the preview could be rendered/calculated. To get the media id, you can single view this broken media see the id in the browser's address bar. With it you check the storage to look for the extracted files. Please provide the directory listing of belonging media directory im the storage.
Further hints can be taken from the internals of the doc.
Hope my comments helps to find answers
xemle
commented
3 years ago Due inactivity I close this ticket. If required please open this ticket.
Thank you very much for your valuable input and feedback.
Hi
I successfully get it running (through binary). Some comments:
I LIKE:
WHAT I MISS/DISLIKE (this is my opinion and to make things better, not criticism):
My 2 cents :)