cvhcikit / sloth

Sloth is a tool for labeling image and video data for computer vision research.

sloth

sloth is a tool for labeling image and video data for computer vision research.

The documentation can be found at http://sloth.readthedocs.org/ .

Latest Releases

2013/11/29 v1.0: 2e69fdae40f89050fbaeef22491eee2a92e78b4f .zip .tar.gz

For a full list, visit https://github.com/cvhciKIT/sloth/releases

sloth's Issues

Make attributes editable via buttons

Move shortcuts to own menu

Saving does not work anymore

• from the gui -> saveAs --> AttributeError: 'MainWindow' object has no attribute 'saveAnnotations'
• using subcommand convert --> Writing to file test.txt failed 'annotations'

Make attributes editable via model tree

It needs to be defined, which attributes are editable and how the editing should work (drop-down box, etc.)
Also, the allowed values need to be stored somewhere where the model (or the TreeView?) can check them.

text not cleared if item with text is deleted and text > boundingRect

cursor in annotations tree jumps to a wrong point on selection after changing labels

After modifying some label points and then clicking on another file in the annotations tree the cursor does not jump to the selected file but to some point at the beginning of the tree.

Selecting an item in the tree view should also select it in the AnnotationScene

Find sane way to sync SceneItems and ModelItems

Currently state is redundant and things are complicated because changes can come from several sources:

• Model.dataChanged -> AnnotationScene.dataChanged -> SceneItem.dataChanged
• SceneItem.itemChanged
• internally -> onKeyPressEvent etc.

Use PIL or other standard python library for loading images

In order to reduce dependencies to OKAPI

On closing, sloth always asks if it should save changes

... even if no changes were made.

Don't add single values to the buttonarea choices

Add fit-to-window zoom mode

Implement search for images/annotations

E.g. search for images that have an annotation of class "face" where and where the attribute "gender" is "male"

implement polygonitem

Show image resolution and current zoom level

Save current frame for the last opened label file

when sloth is restarted, go directly to the last viewed frame

Copy annotation boxes across frames

Along with the current copying annotation labels (ids), it will be nice to give an option (different key binding) to copy the boxes too. Useful if there is not much motion in the video.

Add labeled/unlabeled status field and display it

Should be set unset by a specific shortcut
Text field
Special value "unlabeled"

Rubberband selection clears current selection even with key modifiers

Qt bug: see http://bugreports.qt.nokia.com/browse/QTBUG-6523

Make attribute values extendable at runtime

Possible values should be automatically determined when loading the labels

deleting keys from the dictionary actually cripples the dict in the container

remove test/debug code from frameviewer

Fix Window menu

Make rectangles resizable via mouse

Make button widget scrollable vertically

Annotation/Image properties are ordered randomly

Handle issue of smaller items being stuck behind larger ones

New style configuration

LABELS = (
        {
            'attributes': {
                'class':     'Face',
                'quality':   QUALITY_VALUES,
                'pan':       float,
                'gender':    GENDER_VALUES,
                'ethnicity': ETHNICITY_VALUES,
                'age':       [10, 20, float],
                'id':        ID_VALUES + ['*', '+'],
                'score':     '-1',
            },
            'item':     VidemoFaceBoundingBox,
            'inserter': 'sloth.items.PointItemInserter',
            'hotkey':   'r',
        },
        {
            'attributes': {
                'class':   'Facial Feature',
                'feature': FEATURE_TYPES,
            },
            'inserter': 'sloth.items.PointItemInserter',
            'item':     VidemoFacialFeature,
            'hotkey':   'f',
        },
        {
            'attributes': {
                'class': 'Gender',
                'type':  'tag',
                'value': GENDER_VALUES,
            },
            'item':     'sloth.items.TagItem',
            'inserter': 'sloth.items.TagItemInserter',
        }
)

Maybe use choices to directly specify hotkeys, e.g."'gender': [ 'male [m]', 'female [f]' ]?

Ja, das hatte ich mir auch schon überlegt, allerdings würde ich das mit tuplen machen:

'gender': [ ('male', 'm'), ('female', 'f'), 'unlabeled' ]

Wenns kein tuple ist, dann hat der Eintrag einfach keinen Hotkey.

Fix face detector plugin

Make Items selectable via model tree

Update the documentation

remove annotation state from default annotation container

• change load to return the annotations
• add annotation parameter to save

(see #3 #3 (comment))

Inserters changes

• Move index into constructor
• Add slots to listen for important scene events (image changed)
• finished signal
• Pass model to inserters
• "Labeling XYZ" message

Add a way to configure plugins

implement FeretContainer.serializeToFile()

update documentation (containers.rst) as well, remove "no write support implemented yet"

Fix resize issues with FloatingLayout

Minimum height must be set somehow.

Make videos work and work efficiently

Needs some changes in OKAPI regarding the index of videos

Implement Undo

Make multiple items selectable by rubberband

rect item after resizing + moving problem

1. resize using keys
2. move rect with mouse

--> back to old size

Add global hotkeys for arbitrary commands

Ensure the annotations are clearly visible

Different color would be good. It can still be occluded by other annotations...
Problem with background of same color

could you put the keys z and x for back and forward, and d for delete?

Display popup dialog when loading of annotations failed

Allow numbers as labels

Implement filtering of annotations

E.g. only show annotations of class "face" where the attribute "gender" is "male".

This should then hide all non-matching images/frames/videos in the tree view and the skipping should work accordingly.

implement polygoniteminserter

CLI interface cannot be used with -c

sloth -c /home/mfischer/src/kit/sloth/videmo_config_new_face.py help 2011-08-16 15:07:47,535 CRITICAL sloth.core.labeltool Error loading annotations: No container registered for filename help Traceback (most recent call last): File "/home/mfischer/src/kit/sloth/sloth/bin/sloth", line 14, in <module> labeltool.execute_from_commandline(sys.argv) File "/home/mfischer/src/kit/sloth/sloth/core/labeltool.py", line 156, in execute_from_commandline self.loadAnnotations(args[1], handleErrors=False) File "/home/mfischer/src/kit/sloth/sloth/core/labeltool.py", line 226, in loadAnnotations self._container = self._container_factory.create(fname) File "/home/mfischer/src/kit/sloth/sloth/annotations/container.py", line 71, in create "No container registered for filename %s" % filename sloth.core.exceptions.ImproperlyConfigured: No container registered for filename help

Make resizing behaviour of button box and display widget less insane

rename container

see issue #3 #3 (comment)