eScriptorium issues
https://gitlab.inria.fr/scripta/escriptorium/-/issues
2021-02-24T14:58:16+01:00
https://gitlab.inria.fr/scripta/escriptorium/-/issues/328
Reattribute numbers to line using text regions
2021-02-24T14:58:16+01:00
Thibault Clérice
Reattribute numbers to line using text regions
**Description**
I have been playing around with copy-pasting content into eScriptorium in order to prepare our data for CREMMA. We have a lot of content that are not Transkribus-born or eScriptorium born that we'd wish to transfer onto ...
**Description**
I have been playing around with copy-pasting content into eScriptorium in order to prepare our data for CREMMA. We have a lot of content that are not Transkribus-born or eScriptorium born that we'd wish to transfer onto eScriptorium. To do that, we copy past column by column the content of the text we have.
I made a small video about the whole process : https://www.youtube.com/watch?v=UZy1ajfQH0o
The issue is transcription lines are ordered by vertical appearance and are not computed again when merge into a text region. This lack of functionality is clearly damaging unnecessarily the time we spent transferring the data.
**Description of feature**
Based on text region, renumber globally the lines.
https://gitlab.inria.fr/scripta/escriptorium/-/issues/154
recalculation of polygons does not work immediately
2020-01-24T09:11:15+01:00
daniel stoekl
recalculation of polygons does not work immediately
https://gitlab.inria.fr/scripta/escriptorium/-/issues/16
Recover image processing errors
2019-04-17T19:49:38+02:00
Robin Tissot
Recover image processing errors
In some cases a DocumentPart can be stuck in processing state, it is possible to replicate this by cold shutting down celery.
One idea would be to have a cron run and check the state of parts with a running process state from time to tim...
In some cases a DocumentPart can be stuck in processing state, it is possible to replicate this by cold shutting down celery.
One idea would be to have a cron run and check the state of parts with a running process state from time to time. It doesn't feel that great so there is probably a better option.
https://gitlab.inria.fr/scripta/escriptorium/-/issues/166
red stroke for line beginnings
2020-02-04T18:19:54+01:00
daniel stoekl
red stroke for line beginnings
It seems to me that the angle of the red stroke marking line beginnings reflects mainly the angle of the leftmost segment, not the overall angle of the whole line. Look e.g. here
![image](/uploads/d31aa941be190fdf43fc84cfe4677d01/image.p...
It seems to me that the angle of the red stroke marking line beginnings reflects mainly the angle of the leftmost segment, not the overall angle of the whole line. Look e.g. here
![image](/uploads/d31aa941be190fdf43fc84cfe4677d01/image.png)
![image](/uploads/f712e046a4619b3f99e0e849916ea840/image.png)
![image](/uploads/964dd16a44b975fa9a022107438e2e1a/image.png)
https://gitlab.inria.fr/scripta/escriptorium/-/issues/142
reduce panelization to 2 panels
2020-03-04T16:11:31+01:00
daniel stoekl
reduce panelization to 2 panels
As discussed, we only need two panels, one for the image and one for the transcription (that can be wysiwyg (Robin's) or straight (Hassan's). Since the layout segmentation interface can zoom we do not need the clean page interface anymor...
As discussed, we only need two panels, one for the image and one for the transcription (that can be wysiwyg (Robin's) or straight (Hassan's). Since the layout segmentation interface can zoom we do not need the clean page interface anymore. Neither do we need a separate panel for bw. However, we do need buttons that allow the user to overlay (or not) combinations of:
a) the type of image (color, grayscale, bw)
b) the baselines
c) the boundary polygons masks
d) the regions
e) the reading order indications
https://gitlab.inria.fr/scripta/escriptorium/-/issues/212
regions cannot be deleted
2020-03-18T14:50:16+01:00
daniel stoekl
regions cannot be deleted
I have a case of unsuppressable lines here: document/340/part/26883/edit/
You can play with this image but please do not change any other annotation.
I have a case of unsuppressable lines here: document/340/part/26883/edit/
You can play with this image but please do not change any other annotation.
https://gitlab.inria.fr/scripta/escriptorium/-/issues/326
region scissor mode should disenable point moving
2021-03-04T15:51:33+01:00
daniel stoekl
region scissor mode should disenable point moving
if one is in region scissor mode, the cursor should not be able to move points. sometimes one needs to cut close to points and it is hard to get them without selecting them. for moving points one could simply switch back to non-scissor m...
if one is in region scissor mode, the cursor should not be able to move points. sometimes one needs to cut close to points and it is hard to get them without selecting them. for moving points one could simply switch back to non-scissor mode.
https://gitlab.inria.fr/scripta/escriptorium/-/issues/17
Replace big thumbnail with full size gracefully
2019-12-03T12:21:43+01:00
Robin Tissot
Replace big thumbnail with full size gracefully
In the binarize/segment and transcription UI, load the full size image in the background and replace the thumbnail when it's ready.
Beware that just switching the src attribute will make the image flicker if its really big so find a wa...
In the binarize/segment and transcription UI, load the full size image in the background and replace the thumbnail when it's ready.
Beware that just switching the src attribute will make the image flicker if its really big so find a way to avoid that (using background first ?).
https://gitlab.inria.fr/scripta/escriptorium/-/issues/279
Repolygonisation after XML import
2020-06-23T16:32:29+02:00
Alix Chagué
Repolygonisation after XML import
Is it possible/planned that when importing segmentation from ALTO or PAGE XML, eScriptorium would allow users to trigger the repolygonisation of the lines based on the imported baselines?
For example, part of our data was annotated manu...
Is it possible/planned that when importing segmentation from ALTO or PAGE XML, eScriptorium would allow users to trigger the repolygonisation of the lines based on the imported baselines?
For example, part of our data was annotated manually with Transkribus, so the shape of the lines associated with the baselines is reduced to a simple rectangular box. I would like to keep the baselines but ignore the lines' coordinates and trigger the "segmentation is ready for masks calculation!" button, or something similar.
https://gitlab.inria.fr/scripta/escriptorium/-/issues/197
repolygonisation error
2020-03-04T10:39:43+01:00
daniel stoekl
repolygonisation error
Repolygonisation does not work. It opens an "object:" warning and then breaks the system with a 504 error. This seems to me an urgent error to fix before Monday. If possible create another instance of this button next to the segment butt...
Repolygonisation does not work. It opens an "object:" warning and then breaks the system with a 504 error. This seems to me an urgent error to fix before Monday. If possible create another instance of this button next to the segment button to be able to apply repolygonisation to a selection of pages.
I think it should be possible to repolygonize only a specific line to save calculation costs.
https://gitlab.inria.fr/scripta/escriptorium/-/issues/63
RTL display error in transcription box
2019-05-02T17:45:36+02:00
STOKL BEN EZRA Daniel
RTL display error in transcription box
Both slashes that should be at the left end of the line are displayed wrongly at its right. They are rightly on the end of the line in the transcription text if I copy the text from there and insert it into matlab.
![image](/uploads/50d...
Both slashes that should be at the left end of the line are displayed wrongly at its right. They are rightly on the end of the line in the transcription text if I copy the text from there and insert it into matlab.
![image](/uploads/50d3f44ebe2dbec908c03f670cae2b97/image.png)
https://gitlab.inria.fr/scripta/escriptorium/-/issues/129
Segmentation annotation bug
2020-01-14T14:43:54+01:00
daniel stoekl
Segmentation annotation bug
Sometimes (sorry to be imprecise but I did not understand the WHEN), one cannot draw more lines or drag more points.
E.g. after merging two lines. Then clicking on another two lines (3 and 4). Without unselecting the previous ones by err...
Sometimes (sorry to be imprecise but I did not understand the WHEN), one cannot draw more lines or drag more points.
E.g. after merging two lines. Then clicking on another two lines (3 and 4). Without unselecting the previous ones by error... Merging them. clicking ctrl z - no more drawing possible. Ctrl z deleted also lines 3 and 4.
https://gitlab.inria.fr/scripta/escriptorium/-/issues/123
segmentation annotation bugs
2020-01-14T14:43:22+01:00
daniel stoekl
segmentation annotation bugs
Since today switching to the develop branch, I had more bugs when interacting with the segmentation annotation/correction. Frequently I could not add new lines anymore or delete anymore. F5 reload would help, but that is not a good worka...
Since today switching to the develop branch, I had more bugs when interacting with the segmentation annotation/correction. Frequently I could not add new lines anymore or delete anymore. F5 reload would help, but that is not a good workaround.
It has become considerably more difficult to grab the points of a line for some reason. Perhaps this depends on the width of the line??? Sometimes the cursor would not correspond to the actual coordinates of the cursor. This was visible when trying to draw the second point of a line. Especially trying to draw new lines or move points at the edges of images was difficult.
https://gitlab.inria.fr/scripta/escriptorium/-/issues/119
Segmentation bugs
2020-01-21T08:35:19+01:00
Robin Tissot
Segmentation bugs
Create separate issues as needed:
* ~~ctrl + move doesn't follow the mouse, when moving more than one point they get clamped up.~~
* merging lines more than once create new lines.
* ~~the sign direction should be more visible especially ...
Create separate issues as needed:
* ~~ctrl + move doesn't follow the mouse, when moving more than one point they get clamped up.~~
* merging lines more than once create new lines.
* ~~the sign direction should be more visible especially on high resolution images.~~
* Sometimes when joining lines, one line seems to become shorter (result of two differing line directions???). You seem to be aware of the problem. In fact, the lines are joined and one of the lines is UNDER the other, seemingly going back. This is hard to spot because dots are on the same coordinates. One needs to move one of the points to see that there is actually sthg below.
* ~~use text direction instead of line direction (need to detect it)~~
* ~~Sometimes baselines are indicated twice, thrice or more for the same line for lines that were corrected. This is only shown when leaving the page and going back~~
* Lines are getting duplicated (when cutting/merging?)
https://gitlab.inria.fr/scripta/escriptorium/-/issues/83
Segmentation v2
2019-11-18T11:22:37+01:00
Robin Tissot
Segmentation v2
List of wanted features and improvements:
* draw baselines -> derive polygons
* multi selection
* easier line creation
double clicking is a pain for the hand, maybe one click on empty space/region should start line creation but then how ...
List of wanted features and improvements:
* draw baselines -> derive polygons
* multi selection
* easier line creation
double clicking is a pain for the hand, maybe one click on empty space/region should start line creation but then how do we select a region? Also fix the bug of new line location when zoomed.
* zoom in transcription panel? We could disable selecting lines while dragging, not sure if it would work.
What about colors?
https://gitlab.inria.fr/scripta/escriptorium/-/issues/76
segment manually but transcribe automatically
2019-05-13T12:07:08+02:00
STOKL BEN EZRA Daniel
segment manually but transcribe automatically
I just uploaded a binarized image, segmented manually two lines and wanted to run the transcriber through it. This turned out to be impossible. The system automatically rebinarized, resegmented (completely wrong) and then the transcripti...
I just uploaded a binarized image, segmented manually two lines and wanted to run the transcriber through it. This turned out to be impossible. The system automatically rebinarized, resegmented (completely wrong) and then the transcription is crap of course.
https://gitlab.inria.fr/scripta/escriptorium/-/issues/217
Segtrain
2020-05-05T08:37:01+02:00
daniel stoekl
Segtrain
What exactly is happening in segtrain? I tried to train a model from Scratch on roughly 300 images. The model page gives 66% as a result after 11 epochs and stops there. And the results are terrible.
![image](/uploads/ef5c95b0732e8f2bbec...
What exactly is happening in segtrain? I tried to train a model from Scratch on roughly 300 images. The model page gives 66% as a result after 11 epochs and stops there. And the results are terrible.
![image](/uploads/ef5c95b0732e8f2bbeca0e866d6400f5/image.png)
I train 11 epochs with the same material on the gpu and the results are ok, not fantastic but workable:
![image](/uploads/670498493a8c1a5736e348548f4e4851/image.png)
1. We know that the automatic evaluation does not correspond to manual evaluation. Could we rather do a fix number of epochs and save every 10th or every 50th model as Ben and I are doing on our gpus?
2. Put as little as possible into the testset in order to keep the trainingset as big as possible.
https://gitlab.inria.fr/scripta/escriptorium/-/issues/346
Server not updating from web app.
2021-04-28T11:40:52+02:00
Hayim Lapin
Server not updating from web app.
Firefox on Windows.
* Not recalculating masks unless individual line is double clicked
* Not saving manual resegmentation
Could this just be a load issue on the server end?
Edit: also not renumbering lines after associating with regions.
Firefox on Windows.
* Not recalculating masks unless individual line is double clicked
* Not saving manual resegmentation
Could this just be a load issue on the server end?
Edit: also not renumbering lines after associating with regions.
https://gitlab.inria.fr/scripta/escriptorium/-/issues/13
Share to users
2019-03-01T20:05:11+01:00
Robin Tissot
Share to users
Free form input with users from the same teams as data, we don't want to auto-complete with all users to avoid being able to fetch the whole db of emails.
Free form input with users from the same teams as data, we don't want to auto-complete with all users to avoid being able to fetch the whole db of emails.
https://gitlab.inria.fr/scripta/escriptorium/-/issues/147
shift select then merge problem
2020-01-28T16:57:10+01:00
daniel stoekl
shift select then merge problem
If you shift select several lines and then try to merge them the lines get deleted.
If you shift select several lines and then try to merge them the lines get deleted.