It seems that the when images are resized to fit into the window, the scale factor for the annotations sometimes does not match the image scale factor, so the boxes appear in the wrong place and are the wrong size. I have noticed when the window is resized then the scale factor is recomputed and then the boxes go to the correct location.
A simple solution to this is to add a call to calculateImageScale() on line 288 of imagetagger/imagetagger/annotations/static/js/annotations.js so that the image scale is always computed right before drawing the boxes.
I have seen this problem occur in Firefox and Chrome. I am working with large images which are scaled down to fit in the window.

Team Rhoban has a very simple verification view, on which the user only has to select the images containing balls. This has the advantage that they can give it to any person and this person can do the job without prior knowledge. Furthermore this is also simple to do on a mobile device, e.g. smart phone.
A very simple verification view which shows only the area around the label together with the label and a yes/no button would be good. This would then also be possible to run on a mobile platform.

Right now, the import format is fixed and not customizable like the export format. A custom regular expression could solve this issue, especially when it is automatically provided with every new export format.

In an image set an option should be given to annotate all the images in the set with "Not in Image" of a given type. This would help to handle negative data sets. The feature is very powerful, so only team admins and the creator of the image set should be allowed to use it. Creating that high amount of labels at once generates too many points for a user. As a solution i'd propose using Null as user. The user creating the annotation verifies them automatically (maybe a checkbox has to be set to validate them all).

It would be great to be able to expand or decrease the size of the bounding box with hotkeys similar to the ones moving the bounding box.

When you have the checkbox for "not in the image" and "keep selection for next" checked at the same time, the existing annotations below the pictures will be stuck on "Loading...".
They will load again after reloading the page or unselecting "keep selection for next" and loading the next image.
Not in image also does not keep selected, even though it seems like it should with keep selection.

If you visit the page of any team you will be shown a "Create imageset" field. Actually creating one will show the user an error, but this is confusing for users and should only be shown on team pages where the user has the right to create imagesets.

If a filter for say robot is activated in annotation view as missing annotation, an image is removed from the list as the first robot is annotated. So you can't annotate multiple robots in a single image successively.

Maybe only update the filtered images if you explicitly change the selected image (last, next, back, skip, click image link).

Verification view should also get the image preloading feature.

Some statistics on Global/Team/User scale, to get a better overview about the progress. I'd plot the results with flot.

It would be more intuitive if an imageset could be created on the home page below the existing imagesets. The create form would only require an additional dropdown input for choosing the corresponding team.

optionally zoom into image section when hovering to increase accuracy

When the images gets uploaded as a bulk (not in a .zip folder), the image size is set by default to 800x600. The actual image size won't get measured on upload.

The creator of a set and team admins of the owning team should be able to delete images from imagesets

#! /bin/bash

echo "Please enter Username and Password for imagetagger.mafiasi.de. This script will download into the current directory"
read -p "User: " USER
read -s -p "Password: " PASSWD

BASEURL="https://imagetagger.bit-bots.de"
TMP=mktemp -d imagetagger-XXXX

echo ""

Login

CSRF=wget -nv --keep-session-cookies --save-cookies ${TMP}/cookie ${BASEURL} -O - | grep csrfmiddlewaretoken | tail -n 1 | sed -E "s/.*value='(.*)' .*/\1/g"
wget -nv --referer ${BASEURL}/users/login/ --keep-session-cookies --save-cookies ${TMP}/cookie --load-cookies ${TMP}/cookie --post-data "username=${USER}&password=${PASSWD}&csrfmiddlewaretoken=$CSRF&login=Log in" $BASEURL/users/login/ -O - >/dev/null
if grep --quiet sessionid ${TMP}/cookie; then
echo "Login success"
else
echo "Login failure"
exit 1
fi

Download images

images=wget -nv --keep-session-cookies --save-cookies ${TMP}/cookie --load-cookies ${TMP}/cookie ${BASEURL}/images/imagelist/$1/ -O -
for image in ${images}
do
if ! wget -nv --keep-session-cookies --save-cookies ${TMP}/cookie --load-cookies ${TMP}/cookie --trust-server-names --content-disposition ${BASEURL}${image}; then
echo "Download Failed"
exit 2
fi
done

Some teams seem to be interested in having the possibility to forbid other from changing labels, thus making some sort of "read only" datasets.
This would need some sort of rights management for datasets. Probably, just an option for everybody can change/add labels and only team members can do it, is enough.

Some meta is very interesting for some approaches of computer vision
Most importantly and wide spread is the camera angle and height at the moment of recording.
Furthermore the aperture of the camera as well as maybe some camera calibration values would be interesting
These values should be stored in a standardized format (e.g. euler angles with a defined 0° position) on the server but it has to be possible to export them into different values (e.g. other orientations, or representation as quaternion)
Additionally, there should be the possibility to attach arbitrary meta data to images and image sets for information which go beyond camera angle and height.

It would be good to have an instruction/help page which shows the user how to label objects.
This could also automatically be displayed when the user first logs in

A button in the menu bar (maybe between tools and user) that opens a dropdown list of all the user's teams for quickly reaching the team's site

Annotation drawings don't get cleared on image change.

EDIT: I'm already working on this.

I notice that placeholders in Create new Exportformat page isn't sufficient to create a PASCAL-VOC-like export format.

For example, PASCAL VOC format looks like this:

<annotation>
	<folder>demo</folder>
	<filename>demo3.jpg</filename>
	<path>/home/yuthon/Workspace/demo/demo3.jpg</path>
	<source>
		<database>Unknown</database>
	</source>
	<size>
		<width>799</width>
		<height>401</height>
		<depth>3</depth>
	</size>
	<segmented>0</segmented>
	<object>
		<name>class-1</name>
		<pose>Unspecified</pose>
		<truncated>0</truncated>
		<difficult>0</difficult>
		<bndbox>
			<xmin>487</xmin>
			<ymin>107</ymin>
			<xmax>790</xmax>
			<ymax>318</ymax>
		</bndbox>
	</object>
	<object>
		<name>type-f</name>
		<pose>Unspecified</pose>
		<truncated>0</truncated>
		<difficult>0</difficult>
		<bndbox>
			<xmin>141</xmin>
			<ymin>68</ymin>
			<xmax>363</xmax>
			<ymax>334</ymax>
		</bndbox>
	</object>
</annotation>

The filename and path part isn't shown on given placeholders, thus it's impossible to create a PASCAL-VOC-like export format.

I'll appreciate it if you could supply more placeholders, or directly support built-in PASCAL-VOC-like export format.

in the annotation view, long image names look extremely ugly. It would help to shorten them or to add a horizontal scroll bar to the image list. The itemisation points take a lot of the space and are unnecessary.

When exporting annotations the indentation of each line is not preserved. This breaks formats that rely on indentation to work (like YAML).

For example, a format like this:

# Base format
imageset_%%imageset:
    description: %%setdecsription
    team: %%team
    location: %%setlocation
    %%content

# Image format
    %%imagename:
        dimensions: [%%imagewidth, %%imageheight]
        num_annotations: %%annoamount
        %%annotations

Produces an output like this:

imageset_bitbots-nagoya-01:
    description: %%setdecsription
    team: Bit-Bots
    location: nagoya
    
ball_move_1_00001.png:
        dimensions: [800, 600]
        num_annotations: 5

Here, the issue is that ball_move_1_00001.png: should be indented to line up with location. The same issue occurs with the annotation formats too.

It also appears as though an extra newline is appended to the base format. While this shouldn't break YAML syntax, it is an unexpected occurrence.

Deletes team automatically when the last user leaves the team. Imagesets still owned by the team should get set as public.

For the moment you can create export formats, but not edit them. So when you make an error you have to start over again :/
Some possibility for editing, or at least cloning export formats would be very nice :D

It should be possible to delete an imageset.

Exporting labels in (for example) "AI Ball" format should only include ball labels, nothing else. However, some exports showed goalpost "filename|not_in_image" labels in the export.txt. Tested on two different imagesets now and also re-created the exports -> same behavior.

Most likely the labels are not filtered correctly. Maybe it is just "not_in_image" labels that are not filtered.

Fix should have high priority imho, since this highly affects the usability of exports.

As tagging is for itself very fast, the loading time for the image is relatively long. It would be nice to preload images, so that when going to the next page the image is already in the browser cache.

Loading the image of the next page with javascript or in a hidden object should do this.

If the ImageTagger is used frequently teams will have several datasets making some kind of sorting an importing feature. A first step would be to include a tabbed layout, e.g. "default", "date created", "by size", "test sets", where the default behavior is retained with the "default" tab and the others sort image sets by date of creation, number of images and "test*" matching the beginning of a dataset's name respectively.

Add a tool section in the top right corner, where people can download the scripts (e.g. for rosbag transformation). Each script should have a small explanation.

Currently the file name for label exports is always autogenerated. It would be nice to be able to set this also in the export format, e.g. to make something like "%%imageset.yaml".

It would be nice to be able as admin to send announcements (new version/feature/imageset etc.) to all users. These could be displayed in the home screen. A global admin would send the announcements to all users while the team-specific announcements just get sent to all the team members.

If the annotations are drawn on the image, you can't start a new annotation from within an existing rectangle.

Imagesets could have priorities (i'd suggest high, normal and low) in which they get ordered in the user's home view and a little lable when they are listed elsewhere. So it would be easier to direct the focus of the collaborators towards the most important sets.

It is not clear how goals and lines are to be labelled.
For goals the posts and bars can be labelled. But using a bounding box is not optimal since the shape of the object can be askew. So a different labelling would be necessary, e.g. bottom and top center point together with width.
For the lines, it is possible to label arbitrary line section or intersections (T, X, L) and special points (penalty and center point).

settings_base.py should include all the settings which should work in most environments. For example:

• installed apps
• context processors
• middleware
• upload handlers
• etc.

The settings.py are specific to your deployment. Therefore only an example is provided.

Currently the exported files do not contain any information about the tagged label (i.e. ball or concealed ball).
The current columns are
imagename|x1|y1|x2|y2

Changing this might introduce incompatibilities with current software using this format.

Add an option to imagesets to select a main annotation type. This annotation type will be the preselected annotation type when a user starts to label images of the set.

It would be great to be able to switch between types of annotations (ball, goalpost etc.) by pressing a hotkey e.g. 1-9.

Hello,

I'd like to propose a new feature to the Imagetagger.
It would be very useful a option for uploading previously made markings.
Of course, these old markings has to be in the same standards used by the Imagetagger.

Best regards

Monitoring the number of labels for a specific type (e.g. ball) on a specific data set is quite difficult at the moment. The annotation stats displayed at a data set view should be grouped by type of label.

For ease of use with a high number of datasets we need a fast & simple method to sort & filter datasets. One convenient way would be the implementation of tags/labels for datasets like GitHub for example does with issues. This way admins could create labels and each dataset could be assigned with any number of labels, like "ball", "nagoya", "2017", "eu-ball", "test-set". In order to easily identify labels they should at least have colors (like in GitHub) or even small icons.

Hello, I git clone this repository and install the required dependencies. But when I type "python3 manage.py runserver 134.100.14.61:5432" on my terminal, I get "Is the server running on host "134.100.14.61" and accepting TCP/IP connections on port 5432?" Is there a user-guide for who want to try this tool?

"Concealed" should be an annotation attribute with a checkbox (like "not in image").
This would permit users to

• filter for annotations like "ball" and concealed balls would also be filtered
• annotate concealed balls easier

When adding new annotation types, the "concealed" checkbox should be optional.

It would be nice to have a tooltip providing information about the image lock option.

The team view should get a tabbed layout to add an all time and 30 day annotation high score view (all team members sorted by annotation high score).

Some permission checks just return a 403 error when the user has not the required permissions. This is not really user friendly and should be replaced by a useful redirect and a message.

As the data format for the export file is still evolving, new lines may need to be included into the export file format. The data format, however, is in principle a CSV file format and as such supported by common libraries, such as Pandas for Python etc. To continue to support those libraries and make parsing of the file not harder than it should be, lines which are neither data nor header, should be commented out with a single comment character at the beginning of the line.

For example:

previously the (BitBots) data format looked like this:

Export of Imageset <imagesetname> (ball annotations in bounding boxes)
imagename|x1|y1|x2|y2
FullPic0.jpg|322|308|396|359

the format has changed and now includes an additional line:

Export of Imageset <imagesetname> (ball annotations in bounding boxes)
set[<imagesetname>]
imagename|x1|y1|x2|y2
aufnahme9_FullPic.jpg|296|131|441|282
...

In my oppinion it would be better to mark the first two lines (and possibly more in the future) with a comment char like this:

# Export of Imageset <imagesetname> (ball annotations in bounding boxes)
# set[<imagesetname>]
imagename|x1|y1|x2|y2
aufnahme9_FullPic.jpg|296|131|441|282
...

This should enable common libraries to ignore those lines, and there is no need to write a parser which first finds the first line, which is not part of the header.

For ease of use it should be possible to filter user annotation stats ("Explore -> Users" view) per team.