Developing Pipelines¶

LOST Ecosystem¶

All about Pipelines¶

#TODO JJ Review

Pipeline Projects¶

A pipeline project in LOST is defined as a folder or git repository that contains pipeline definition files in json format and related python3 scripts. Additional, other files can be placed into this folder that can be accessed by the scripts of a pipeline.

Pipeline Project Examples¶

Pipeline project examples can be found here: LOST Out of the Box Pipelines

Repository Structure¶

Example repo structure of a lost pipeline project.¶

lost_ootb_pipes/
├── found
│   ├── cluster_using_prev_stage.py
│   ├── __init__.py
│   ├── mia.json
│   ├── mia_request_again.json
│   ├── mia_sia.json
│   ├── request_annos_again.py
│   ├── request_annos.py
│   ├── request_images_by_lbl.py
│   ├── sia.json
│   ├── sia_request_again.json
│   └── two_stage.json
├── LICENSE
└── README.md

1 directory, 13 files

The listing above shows an example for a pipeline directory structure. Where the root folder lost_ootb_pipes is the repo name and found is the name of the pipeline project. found contains all files required for the pipelines of the pipeline project. Within the project there are json files where each represents a pipeline definition. A pipeline is composed from different scripts (request_annos.py, request_annos_again.py, request_images._by_lbl.py) and other pipeline elements. Some of the scripts may require a special python package you have written. So if you want to use this package (e.g. my_special_python_lib), just place it also inside the pipeline project folder. Sometimes it is also useful to place some files into the project folder, for example a pretrained ai model that should be loaded inside a script.

Importing a Pipeline Project into LOST¶

After creating a pipeline it needs to be imported into LOST. In order to do that, you need to perform the following steps:

Log into LOST as Admin
Go to Admin Area
Click on the Pipeline Projects tab
Click on Import pipeline project button
Click on Import/ Update pipeline project from a public git repository
Add the url of the pipeline project you like to import
Click on Import/ Update

Pipeline import GUI

Updating a LOST Pipeline¶

If there was an update for one of your pipelines you need to update your pipe project in LOST.In order to do this, the procedure is the same as for importing a pipeline

Namespacing¶

When importing or updating a pipeline project in LOST the following namespacing will be applied to pipelines: <name of pipeline project folder>.<name of pipeline json file>. In the same way scripts will be namespaced internally by LOST: <name of pipeline project folder>.<name of python script file>.

So in our example the pipelines would be named found.mia and found.mia_request_again ….

Pipeline Definition Files¶

Within the pipeline definition file you define your annotation process. Such a pipeline is composed of different standard elements that are supported by LOST like datasource, script, annotTask, dataExport, visualOutput and loop. Each pipeline element is represented by a json object inside the pipeline definition.

As you can see in the example, the pipeline itself is also defined by a json object. This object has a description, a author, a pipe-schema-version and a list of pipeline elements. Each element object has a peN (pipeline element number) which is the identifier of the element itself. An element needs also an attribute that is called peOut and contains a list of elements where the current element is connected to.

An Example¶

Possible Pipeline Elements¶

Below you will find the definition of all possible pipeline elements in LOST.

 {
   "peN" : "[int]",
   "peOut" : "[list of int]|[null]",
   "datasource" : {
     "type" : "rawFile"
   }
 }

Datasource elements are intended to provide datasets to Script elements. To be more specific it will provide a path inside the LOST system. In most cases this will be a path to a folder with images that should be annotated. The listing above shows the definition of a Datasource element. At the current state only type rawFile is supported, which will provide a path.

 {
   "peN" : "[int]",
   "peOut" : "[list of int]|[null]",
   "script" : {
     "path": "[string]",
     "description" : "[string]"
   }
 }

Script elements represent python3 scripts that are executed as part of your pipeline. In order to define a Script you need to specify a path to the script file relative to the pipeline project folder and a short description of your script.

 {
   "peN" : "[int]",
   "peOut" : "[list of int]|[null]",
   "annoTask" : {
     "type" : "mia|sia",
     "name" : "[string]",
     "instructions" : "[string]",
     "configuration":{"..."}
   }
 }

An AnnoTask represents an annotation task for a human-in-the-loop. Scripts can request annotations for specific images that will be presented in one of the annotation tools in the web gui.

Right now two types of annotation tools are available. If you set type to sia the single image annotation tool will be used for annotation. When choosing mia the images will be present in the multi image annotation tool.

An AnnoTask requires also a name and instructions for the annotator. Based on the type a specific configuration is required.

If “type” is “mia” the configuration will be the following:

 {
   "type": "annoBased|imageBased",
   "showProposedLabel": "[boolean]",
   "drawAnno": "[boolean]",
   "addContext": "[float]"
 }

MIA configuration:

type
- If imageBased a whole image will be presented in the clustered view.
- If annoBased all lost.db.model.TwoDAnno objects related to an image will be cropped and presented in the clustered view.
showProposedLabel
- If true, the assigned sim_class will be interpreted as label and be used as pre-selection of the label in the MIA tool.
drawAnno
- If true and type : annoBased the specific annotation will be drawn inside the cropped image.
addContext
- If type : annoBased and addContext > 0.0, some amount of pixels will be added around the annotation when the annotation is cropped. The number of pixels that are add is calculated relative to the image size. So if you set addContext to 0.1, 10 percent of the image size will be added to the crop. This setting is useful to provide the annotator some more visual context during the annotation step.

If “type” is “sia” the configuration will be the following:

 {
   "tools": {
           "point": "[boolean]",
           "line": "[boolean]",
           "polygon": "[boolean]",
           "bbox": "[boolean]",
           "junk": "[boolean]"
   },
   "annos":{
       "multilabels": "[boolean]",
       "actions": {
           "draw": "[boolean]",
           "label": "[boolean]",
           "edit": "[boolean]",
       },
       "minArea": "[int]",
       "maxAnnos": "[int or null]"
   },
   "img": {
       "multilabels": "[boolean]",
       "actions": {
           "label": "[boolean]",
       }
   }
 }

SIA configuration:

tools
- Inside the tools object you can select which drawing tools are available and if the junk button is present in the SIA gui. You may choose either true or false for each of the tools (point, line, polygon, bbox, junk).
annos (configuration for annotations on the image)
- actions
  
  draw is set to false a user may not draw any new annotations. This is useful if a script sent annotation proposals to SIA and the user should only correct the proposed annotations.
  
  label allows to disable the possibility to assign labels to annotations. This option is useful if you wish that your annotator will only draw annotations.
  
  edit inidcates wether an annotator may edit an annotation that is already present.
- multilabels allows to assign multiple labels per annotation.
- minArea The minimum area in pixels that an annotation may have. This constraint is only applied to annotations where an area can be defined (e.g. BBoxs, Polygons).
- maxAnnos Maximum number of annos that are allowed per image. If null an infinite number of annotation are allowed per image.
img (configuration for the image)
- actions
  
  label allows to disable the possibility to assign labels to the image.
- multilabels allows to assign multiple labels to the image.

 {
   "peN" : "[int]",
   "peOut" : "[list of int]|[null]",
   "dataExport" : {}
 }

A DataExport is used to serve a file generated by a script in the web gui. No special configuration is required for this pipeline element. The file to download will be provided by a Script that is connected to the input of the DataExport element. This Script will call the lost.pyapi.inout.ScriptOutput.add_data_export() method in order to do that.

 {
   "peN" : "[int]",
   "peOut" : "[list of int]|[null]",
   "visualOutput" : {}
 }

A VisualOutput element can display images and html text inside the LOST web gui. A connected Script element will provide the content to an VisualOutput by calling lost.pyapi.inout.ScriptOutput.add_visual_output().

 {
   "peN": "[int]",
   "peOut": "[list of int]|[null]",
   "loop": {
     "maxIteration": "[int]|[null]",
     "peJumpId": "[int]"
   }
 }

A Loop element can be used to build learning loops inside of a pipeline. Such a Loop models a similar behaviour to a while loop in a programming language.

The peJumpId defines the peN of another element in the pipeline where this Loop should jump to while looping. The maxIteration setting inside a loop definition can be set to a maximum amount of iterations that should be performed or to null in order to have an infinity loop.

A Script element inside a loop cycle may break a loop by calling lost.pyapi.script.Script.break_loop(). Scripts inside a loop cycle may check if a loop was broken by calling lost.pyapi.script.Script.loop_is_broken().

All about Scripts¶

pyapi¶

#TODO JJ Review

class lost.pyapi.script.Script(pe_id=None)[source]¶

Superclass for a user defined Script.

Custom scripts need to inherit from Script and implement the main method.

pe_id¶

Pipe element id. Assign the pe id of a pipline script in order to emulate this script in a jupyter notebook for example.

Type:	int

break_loop()[source]¶: Break next loop in pipeline.

create_label_tree(name, external_id=None)[source]¶

Create a new LabelTree

Parameters:	name (str) – Name of the tree / name of the root leaf. external_id (str) – An external id for the root leaf.
Returns:	The created LabelTree.
Return type:	`lost.logic.label.LabelTree`

get_alien_element(pe_id)[source]¶

Get an pipeline element by id from somewhere in the LOST system.

It is an alien element since it is most likely not part of the pipeline instance this script belongs to.

Parameters:	pe_id (int) – PipeElementID of the alien element.
Returns:	`lost.pyapi.script.Script` `lost.pyapi.pipe_elements.AnnoTask` `lost.pyapi.pipe_elements.Datasource` `lost.pyapi.pipe_elements.VisualOutput` `lost.pyapi.pipe_elements.DataExport` `lost.pyapi.pipe_elements.Loop`

get_arg(arg_name)[source]¶

Get argument value by name for this script.

Parameters:	arg_name (str) – Name of the argument.
Returns:	Value of the given argument.

get_fs(name=None)[source]¶

Get default lost filesystem or a specific filesystem by name.

Returns:	See https://filesystem-spec.readthedocs.io/en/latest/api.html#fsspec.spec.AbstractFileSystem
Return type:	fsspec.spec.AbstractFileSystem

get_label_tree(name)[source]¶

Get a LabelTree by name.

Parameters:	name (str) – Name of the desired LabelTree.

Retruns:

lost.logic.label.LabelTree or None:: If a label tree with the given name exists it will be returned. Otherwise None will be returned

get_path(file_name, context='instance')[source]¶

Get path for the filename in a specific context in filesystem.

Parameters:	file_name (str) – Name or relative path for a file. context (str) – Options: instance, pipe
Returns:	Absolute path to the file in the specified context.
Return type:	str

inp¶: lost.pyapi.inout.Input

iteration¶

Get the current iteration.

Number of times this script has been executed.

Type:	int

logger¶

A standard python logger for this script.

It will log to the pipline log file.

Type:	`logging.Logger`

loop_is_broken()[source]¶: Check if the current loop is broken

outp¶: lost.pyapi.inout.ScriptOutput

pipe_info¶

An object with pipeline informations

Type:	`lost.pyapi.pipeline.PipeInfo`

progress¶

Get current progress that is displayed in the progress bar of this script.

Current progress in percent 0…100

Type:	float

reject_execution()[source]¶: Reject execution of this script and set it to PENDING again.

Note

This method is useful if you want to execute this script only when some condition based on previous pipeline elements is meet.

report_err(msg)[source]¶

Report an error for this user script to portal

Parameters:	msg – The error message that should be reported.

Note

You can call this method multiple times if you like. All messages will be concatenated an sent to the portal.

update_progress(value)[source]¶

Update the progress for this script.

Parameters:	value (float) – Progress in percent 0…100

inout¶

ScriptOutput¶

class lost.pyapi.inout.ScriptOutput(script)[source]¶

Special Output class since lost.pyapi.script.Script objects may manipulate and request annotations.

add_data_export(file_path, fs)[source]¶

Serve a file for download inside the web gui via a DataExport element.

Parameters:	file_path (str) – Path to the file that should be provided for download. fs (filesystem) – Filesystem, where file_path is valid.

add_visual_output(img_path=None, html=None)[source]¶

Display an image and html in the web gui via a VisualOutput element.

Parameters:	img_path (str) – Path in the lost filesystem to the image to display. html (str) – HTML text to display.

anno_tasks¶: list of lost.pyapi.pipe_elements.AnnoTask objects

bbox_annos¶

Iterate over all bbox annotation.

Returns:	Iterator of `lost.db.model.TwoDAnno`.

data_exports¶: list of lost.pyapi.pipe_elements.VisualOutput objects.

datasources¶: list of lost.pyapi.pipe_elements.Datasource objects

img_annos¶

Iterate over all lost.db.model.ImageAnno objects in this Resultset.

Returns:	Iterator of `lost.db.model.ImageAnno` objects.

line_annos¶

Iterate over all line annotations.

Returns:	Iterator of `lost.db.model.TwoDAnno` objects.

mia_tasks¶: list of lost.pyapi.pipe_elements.MIATask objects

point_annos¶

Iterate over all point annotations.

Returns:	Iterator of `lost.db.model.TwoDAnno`.

polygon_annos¶

Iterate over all polygon annotations.

Returns:	Iterator of `lost.db.model.TwoDAnno` objects.

request_annos(img, img_labels=None, img_sim_class=None, annos=[], anno_types=[], anno_labels=[], anno_sim_classes=[], frame_n=None, video_path=None, fs=None, img_meta=None, anno_meta=None, img_comment=None)[source]¶

Request annotations for a subsequent annotaiton task.

Parameters:

img (str or ImageAnno) – Path to the image or database image where annotations will be requested for
img_label (list of int) – Labels that will be assigned to the image. The labels should be represented by a label_leaf_id. An image may have multiple labels.
img_sim_class (int) – A culster id that will be used to cluster this image in the MIA annotation tool.
annos (list of list) – A list of POINTs: [x,y] BBOXes: [x,y,w,h] LINEs or POLYGONs: [[x,y], [x,y], …]
anno_types (list of str) – Can be ‘point’, ‘bbox’, ‘line’, ‘polygon’
anno_labels (list of int) – Labels for the twod annos. Each label in the list is represented by a label_leaf_id. (see also LabelLeaf).
anno_sim_classes (list of ints) – List of arbitrary cluster ids that are used to cluster annotations in the MIA annotation tool.
frame_n (int) – If img_path belongs to a video frame_n indicates the framenumber.
video_path (str) – If img_path belongs to a video this is the path to this video.
fs (fsspec.spec.AbstractFileSystem) – The filesystem where image is located. Use lost standard filesystem if no filesystem was given. You can get this Filesystem object from a DataSource-Element by calling get_fm method.
img_meta (dict) – Dictionary with meta information that should be added to the image annotation. Each meta key will be added as column during annotation export. the dict-value will be row content.
anno_meta (list of dict) – List of dictionaries with meta information that should be added to a specific annotation. Each meta key will be added as column during annotation export. The dict-value will be row content.
img_comment (str) – A comment that will be added to this image.

Example

Request human annotations for an image with annotation proposals:

 >>> self.outp.request_annos('path/to/img.jpg',
...     annos = [
...         [0.1, 0.1, 0.2, 0.2],
...         [0.1, 0.2],
...         [[0.1, 0.3], [0.2, 0.3], [0.15, 0.1]]
...     ],
...     anno_types=['bbox', 'point', 'polygon'],
...     anno_labels=[
...         [1],
...         [1],
...         [4]
...     ],
...     anno_sim_classes=[10, 10, 15]
... )

Reqest human annotations for an image without porposals:

>>> self.outp.request_annos('path/to/img.jpg')

request_lds_annos(lds, fs=None, anno_meta_keys=[], img_meta_keys=[], img_path_key='img_path')[source]¶

Request annos from LOSTDataset.

Parameters:

lds (LOSTDataset) – A lost dataset object. Request all annotation in this dataset again.
fs (fsspec.spec.AbstractFileSystem) – The filesystem where image is located. Use lost standard filesystem if no filesystem was given. You can get this Filesystem object from a DataSource-Element by calling get_fm method.
img_meta_keys (list) – Keys that should be used for img_anno meta information
anno_meta_keys (list) – Keys that should be used for two_d_anno meta information

sia_tasks¶: list of lost.pyapi.pipe_elements.SIATask objects

to_df()¶

Get a pandas DataFrame of all annotations related to this object.

Returns:

Column names are:: ’img_uid’, img_timestamp’, img_state’, img_sim_class’, img_frame_n’, ‘img_path’,’img_iteration’,’img_user_id’,’img_anno_time’,’img_lbl’, ‘img_lbl_id’,’img_user’,’img_is_junk’,’img_fs_name’,’anno_uid’, ‘anno_timestamp’,’anno_state’,’anno_dtype’,’anno_sim_class’, ‘anno_iteration’,’anno_user_id’,’anno_user’,’anno_confidence’, ‘anno_time’,’anno_lbl’,’anno_lbl_id’,’anno_style’,’anno_format’, ‘anno_comment’,’anno_data’

Return type: pandas.DataFrame

to_vec(columns='all')¶

Get a vector of all Annotations related to this object.

Parameters: columns (str or list of str) – ‘all’ OR ‘img_uid’, img_timestamp’, img_state’, img_sim_class’, img_frame_n’, ‘img_path’,’img_iteration’,’img_user_id’,’img_anno_time’,’img_lbl’, ‘img_lbl_id’,’img_user’,’img_is_junk’,’img_fs_name’,’anno_uid’, ‘anno_timestamp’,’anno_state’,’anno_dtype’,’anno_sim_class’, ‘anno_iteration’,’anno_user_id’,’anno_user’,’anno_confidence’, ‘anno_time’,’anno_lbl’,’anno_lbl_id’,’anno_style’,’anno_format’, ‘anno_comment’,’anno_data’

Retruns:: list OR list of lists: Desired columns

Example

Return just a list of 2d anno labels:

>>> self.outp.to_vec('anno_lbl')
[['Person'],[],['Cat'],[],['Car'],['Person'],[],['Bird'],['Bird']]

Return a list of lists:

>>> self.outp.to_vec(['img_path', 'anno_lbl',
...     'anno_data', 'anno_dtype'])
[
    ['path/to/img1.jpg', ['Aeroplane'], [[0.1, 0.1, 0.2, 0.2]], 'bbox'],
    ['path/to/img1.jpg', ['Bicycle'], [[0.1, 0.1]], 'point'],
    ['path/to/img2.jpg', ['Bottle'], [[0.1, 0.1], [0.2, 0.2]], 'line'],
    ['path/to/img3.jpg', ['Horse'], [[0.2, 0.15, 0.3, 0.18]], 'bbox']
]

twod_annos¶

Iterate over 2D-annotations.

Returns:	of `lost.db.model.TwoDAnno` objects.
Return type:	Iterator

visual_outputs¶: list of lost.pyapi.pipe_elements.VisualOutput objects.

Output¶

class lost.pyapi.inout.Output(element)[source]¶

anno_tasks¶: list of lost.pyapi.pipe_elements.AnnoTask objects

bbox_annos¶

Iterate over all bbox annotation.

Returns:	Iterator of `lost.db.model.TwoDAnno`.

data_exports¶: list of lost.pyapi.pipe_elements.VisualOutput objects.

datasources¶: list of lost.pyapi.pipe_elements.Datasource objects

img_annos¶

Iterate over all lost.db.model.ImageAnno objects in this Resultset.

Returns:	Iterator of `lost.db.model.ImageAnno` objects.

line_annos¶

Iterate over all line annotations.

Returns:	Iterator of `lost.db.model.TwoDAnno` objects.

mia_tasks¶: list of lost.pyapi.pipe_elements.MIATask objects

point_annos¶

Iterate over all point annotations.

Returns:	Iterator of `lost.db.model.TwoDAnno`.

polygon_annos¶

Iterate over all polygon annotations.

Returns:	Iterator of `lost.db.model.TwoDAnno` objects.

sia_tasks¶: list of lost.pyapi.pipe_elements.SIATask objects

to_df()¶

Get a pandas DataFrame of all annotations related to this object.

Returns:

Column names are:: ’img_uid’, img_timestamp’, img_state’, img_sim_class’, img_frame_n’, ‘img_path’,’img_iteration’,’img_user_id’,’img_anno_time’,’img_lbl’, ‘img_lbl_id’,’img_user’,’img_is_junk’,’img_fs_name’,’anno_uid’, ‘anno_timestamp’,’anno_state’,’anno_dtype’,’anno_sim_class’, ‘anno_iteration’,’anno_user_id’,’anno_user’,’anno_confidence’, ‘anno_time’,’anno_lbl’,’anno_lbl_id’,’anno_style’,’anno_format’, ‘anno_comment’,’anno_data’

Return type: pandas.DataFrame

to_vec(columns='all')¶

Get a vector of all Annotations related to this object.

Parameters: columns (str or list of str) – ‘all’ OR ‘img_uid’, img_timestamp’, img_state’, img_sim_class’, img_frame_n’, ‘img_path’,’img_iteration’,’img_user_id’,’img_anno_time’,’img_lbl’, ‘img_lbl_id’,’img_user’,’img_is_junk’,’img_fs_name’,’anno_uid’, ‘anno_timestamp’,’anno_state’,’anno_dtype’,’anno_sim_class’, ‘anno_iteration’,’anno_user_id’,’anno_user’,’anno_confidence’, ‘anno_time’,’anno_lbl’,’anno_lbl_id’,’anno_style’,’anno_format’, ‘anno_comment’,’anno_data’

Retruns:: list OR list of lists: Desired columns

Example

Return just a list of 2d anno labels:

>>> self.outp.to_vec('anno_lbl')
[['Person'],[],['Cat'],[],['Car'],['Person'],[],['Bird'],['Bird']]

Return a list of lists:

>>> self.outp.to_vec(['img_path', 'anno_lbl',
...     'anno_data', 'anno_dtype'])
[
    ['path/to/img1.jpg', ['Aeroplane'], [[0.1, 0.1, 0.2, 0.2]], 'bbox'],
    ['path/to/img1.jpg', ['Bicycle'], [[0.1, 0.1]], 'point'],
    ['path/to/img2.jpg', ['Bottle'], [[0.1, 0.1], [0.2, 0.2]], 'line'],
    ['path/to/img3.jpg', ['Horse'], [[0.2, 0.15, 0.3, 0.18]], 'bbox']
]

twod_annos¶

Iterate over 2D-annotations.

Returns:	of `lost.db.model.TwoDAnno` objects.
Return type:	Iterator

visual_outputs¶: list of lost.pyapi.pipe_elements.VisualOutput objects.

Input¶

class lost.pyapi.inout.Input(element)[source]¶

Class that represants an input of a pipeline element.

Parameters:	element (object) – Related `lost.db.model.PipeElement` object.

anno_tasks¶: list of lost.pyapi.pipe_elements.AnnoTask objects

bbox_annos¶

Iterate over all bbox annotation.

Returns:	Iterator of `lost.db.model.TwoDAnno`.

data_exports¶: list of lost.pyapi.pipe_elements.VisualOutput objects.

datasources¶: list of lost.pyapi.pipe_elements.Datasource objects

img_annos¶

Iterate over all lost.db.model.ImageAnno objects in this Resultset.

Returns:	Iterator of `lost.db.model.ImageAnno` objects.

line_annos¶

Iterate over all line annotations.

Returns:	Iterator of `lost.db.model.TwoDAnno` objects.

mia_tasks¶: list of lost.pyapi.pipe_elements.MIATask objects

point_annos¶

Iterate over all point annotations.

Returns:	Iterator of `lost.db.model.TwoDAnno`.

polygon_annos¶

Iterate over all polygon annotations.

Returns:	Iterator of `lost.db.model.TwoDAnno` objects.

sia_tasks¶: list of lost.pyapi.pipe_elements.SIATask objects

to_df()[source]¶

Get a pandas DataFrame of all annotations related to this object.

Returns:

Column names are:: ’img_uid’, img_timestamp’, img_state’, img_sim_class’, img_frame_n’, ‘img_path’,’img_iteration’,’img_user_id’,’img_anno_time’,’img_lbl’, ‘img_lbl_id’,’img_user’,’img_is_junk’,’img_fs_name’,’anno_uid’, ‘anno_timestamp’,’anno_state’,’anno_dtype’,’anno_sim_class’, ‘anno_iteration’,’anno_user_id’,’anno_user’,’anno_confidence’, ‘anno_time’,’anno_lbl’,’anno_lbl_id’,’anno_style’,’anno_format’, ‘anno_comment’,’anno_data’

Return type: pandas.DataFrame

to_vec(columns='all')[source]¶

Get a vector of all Annotations related to this object.

Parameters: columns (str or list of str) – ‘all’ OR ‘img_uid’, img_timestamp’, img_state’, img_sim_class’, img_frame_n’, ‘img_path’,’img_iteration’,’img_user_id’,’img_anno_time’,’img_lbl’, ‘img_lbl_id’,’img_user’,’img_is_junk’,’img_fs_name’,’anno_uid’, ‘anno_timestamp’,’anno_state’,’anno_dtype’,’anno_sim_class’, ‘anno_iteration’,’anno_user_id’,’anno_user’,’anno_confidence’, ‘anno_time’,’anno_lbl’,’anno_lbl_id’,’anno_style’,’anno_format’, ‘anno_comment’,’anno_data’

Retruns:: list OR list of lists: Desired columns

Example

Return just a list of 2d anno labels:

>>> self.outp.to_vec('anno_lbl')
[['Person'],[],['Cat'],[],['Car'],['Person'],[],['Bird'],['Bird']]

Return a list of lists:

>>> self.outp.to_vec(['img_path', 'anno_lbl',
...     'anno_data', 'anno_dtype'])
[
    ['path/to/img1.jpg', ['Aeroplane'], [[0.1, 0.1, 0.2, 0.2]], 'bbox'],
    ['path/to/img1.jpg', ['Bicycle'], [[0.1, 0.1]], 'point'],
    ['path/to/img2.jpg', ['Bottle'], [[0.1, 0.1], [0.2, 0.2]], 'line'],
    ['path/to/img3.jpg', ['Horse'], [[0.2, 0.15, 0.3, 0.18]], 'bbox']
]

twod_annos¶

Iterate over 2D-annotations.

Returns:	of `lost.db.model.TwoDAnno` objects.
Return type:	Iterator

visual_outputs¶: list of lost.pyapi.pipe_elements.VisualOutput objects.

pipeline¶

PipeInfo¶

class lost.pyapi.pipeline.PipeInfo(pipe, dbm)[source]¶

description¶

Description that was defined when pipeline was started.

Type:	str

logfile_path¶

Path to pipeline log file.

Type:	str

name¶

Name of this pipeline

Type:	str

timestamp¶

Timestamp when pipeline was started.

Type:	str

timestamp_finished¶

Timestamp when pipeline was finished.

Type:	str

user¶

User who started this pipe

Type:	User object

pipe_elements¶

Datasource¶

class lost.pyapi.pipe_elements.Datasource(pe, dbm)[source]¶

get_fs()[source]¶: Get filesystem for this datasource

inp¶

Input of this pipeline element

Type:	`lost.pyapi.inout.Input`

outp¶

Output of this pipeline element

Type:	`lost.pyapi.inout.Output`

path¶

Relative path to file or folder

Type:	str

pipe_info¶

An object with pipeline informations

Type:	`lost.pyapi.pipeline.PipeInfo`

AnnoTask¶

class lost.pyapi.pipe_elements.AnnoTask(pe, dbm)[source]¶

configuration¶

Configuration of this annotask.

Type:	str

inp¶

Input of this pipeline element

Type:	`lost.pyapi.inout.Input`

instructions¶

Instructions for the annotator of this AnnoTask.

Type:	str

lbl_map¶

Map lbl_name to idx

Note

All label names will be mapped to lower case!

Type:	dict

name¶

A name for this annotask.

Type:	str

outp¶

Output of this pipeline element

Type:	`lost.pyapi.inout.Output`

pipe_info¶

An object with pipeline informations

Type:	`lost.pyapi.pipeline.PipeInfo`

possible_label_df¶

Get all possible labels for this annotation task in DataFrame format

pd.DataFrame: Column names are:: ‘idx’, ‘name’, ‘abbreviation’, ‘description’, ‘timestamp’, ‘external_id’, ‘is_deleted’, ‘parent_leaf_id’ ,’is_root’

Type:	pd.DataFrame

progress¶

Progress in percent.

Value range 0…100.

Type:	float

MIATask¶

class lost.pyapi.pipe_elements.MIATask(pe, dbm)[source]¶

configuration¶

Configuration of this annotask.

Type:	str

inp¶

Input of this pipeline element

Type:	`lost.pyapi.inout.Input`

instructions¶

Instructions for the annotator of this AnnoTask.

Type:	str

lbl_map¶

Map lbl_name to idx

Note

All label names will be mapped to lower case!

Type:	dict

name¶

A name for this annotask.

Type:	str

outp¶

Output of this pipeline element

Type:	`lost.pyapi.inout.Output`

pipe_info¶

An object with pipeline informations

Type:	`lost.pyapi.pipeline.PipeInfo`

possible_label_df¶

Get all possible labels for this annotation task in DataFrame format

pd.DataFrame: Column names are:: ‘idx’, ‘name’, ‘abbreviation’, ‘description’, ‘timestamp’, ‘external_id’, ‘is_deleted’, ‘parent_leaf_id’ ,’is_root’

Type:	pd.DataFrame

progress¶

Progress in percent.

Value range 0…100.

Type:	float

SIATask¶

class lost.pyapi.pipe_elements.SIATask(pe, dbm)[source]¶

configuration¶

Configuration of this annotask.

Type:	str

inp¶

Input of this pipeline element

Type:	`lost.pyapi.inout.Input`

instructions¶

Instructions for the annotator of this AnnoTask.

Type:	str

lbl_map¶

Map lbl_name to idx

Note

All label names will be mapped to lower case!

Type:	dict

name¶

A name for this annotask.

Type:	str

outp¶

Output of this pipeline element

Type:	`lost.pyapi.inout.Output`

pipe_info¶

An object with pipeline informations

Type:	`lost.pyapi.pipeline.PipeInfo`

possible_label_df¶

Get all possible labels for this annotation task in DataFrame format

pd.DataFrame: Column names are:: ‘idx’, ‘name’, ‘abbreviation’, ‘description’, ‘timestamp’, ‘external_id’, ‘is_deleted’, ‘parent_leaf_id’ ,’is_root’

Type:	pd.DataFrame

progress¶

Progress in percent.

Value range 0…100.

Type:	float

DataExport¶

class lost.pyapi.pipe_elements.DataExport(pe, dbm)[source]¶

file_path¶

A list of absolute path to exported files

Type:	list of str

inp¶

Input of this pipeline element

Type:	`lost.pyapi.inout.Input`

outp¶

Output of this pipeline element

Type:	`lost.pyapi.inout.Output`

pipe_info¶

An object with pipeline informations

Type:	`lost.pyapi.pipeline.PipeInfo`

to_dict()[source]¶

Transform a list of exports to a dictionary.

Returns:	[{‘iteration’:int, ‘file_path’:str},…]
Return type:	list of dict

VisualOutput¶

class lost.pyapi.pipe_elements.VisualOutput(pe, dbm)[source]¶

html_strings¶

list of html strings.

Type:	list of str

img_paths¶

List of absolute paths to images.

Type:	list of str

inp¶

Input of this pipeline element

Type:	`lost.pyapi.inout.Input`

outp¶

Output of this pipeline element

Type:	`lost.pyapi.inout.Output`

pipe_info¶

An object with pipeline informations

Type:	`lost.pyapi.pipeline.PipeInfo`

to_dict()[source]¶

Transforms a list of visualization information into a list of dicts.

Returns:	[{‘iteration’:int, ‘img_path’:str, ‘html_string’:str},…]
Return type:	list of dicts

Loop¶

class lost.pyapi.pipe_elements.Loop(pe, dbm)[source]¶

inp¶

Input of this pipeline element

Type:	`lost.pyapi.inout.Input`

is_broken¶

True if loop is broken

Type:	bool

iteration¶

Current iteration of this loop.

Type:	int

max_iteration¶

Maximum number of iteration.

Type:	int

outp¶

Output of this pipeline element

Type:	`lost.pyapi.inout.Output`

pe_jump¶

PipelineElement where this loop will jump to when looping.

Can be of type:

pipe_info¶

An object with pipeline informations

Type:	`lost.pyapi.pipeline.PipeInfo`

model¶

ImageAnno¶

class lost.db.model.ImageAnno(anno_task_id=None, user_id=None, timestamp=None, state=None, sim_class=None, result_id=None, img_path=None, frame_n=None, video_path=None, iteration=0, anno_time=None, is_junk=None, description=None, fs_id=None, meta=None, meta_blob=None, img_actions=None)[source]¶

An ImageAnno represents an image annotation.

Multiple labels as well as 2d annotations (e.g. points, lines, boxes, polygons) can be assigned to an image.

labels¶

The related Label object.

Type:	list

twod_annos¶

A list of TwoDAnno objects.

Type:	list

img_path¶

Abs path to image in file system

Type:	str

frame_n¶

If this image is part of an video, frame_n indicates the frame number.

Type:	int

video_path¶

If this image is part of an video, this should be the path to that video in file system.

Type:	str

sim_class¶

The similarity class this anno belong to. It is used to cluster similar annos in MIA

Type:	int

anno_time¶: Overall annotation time in seconds.

timestamp¶

Timestamp of ImageAnno

Type:	DateTime

iteration¶

The iteration of a loop when this anno was created.

Type:	int

idx¶

ID of this ImageAnno in database

Type:	int

anno_task_id¶

ID of the anno_task this ImageAnno belongs to.

Type:	int

state¶

See lost.db.state.Anno

Type:	enum

result_id¶: Id of the related result.

user_id¶

Id of the annotator.

Type:	int

is_junk¶

This image was marked as Junk.

Type:	bool

description¶

Description for this annotation. Assigned by an annotator or algorithm.

Type:	str

fs_id¶

Id of the filesystem where image is located

Type:	int

meta¶

A field for meta information added by a script

Type:	str

img_actions¶

Actions performed by users for this image

Type:	str

get_anno_vec(anno_type='bbox')[source]¶

Get related 2d annotations in list style.

Parameters:	anno_type (str) – Can be ‘bbox’, ‘point’, ‘line’, ‘polygon’
Returns:	For POINTs: [[x, y], [x, y], …] For BBOXs: [[x, y, w, h], [x, y, w, h], …] For LINEs and POLYGONs: [[[x, y], [x, y],…], [[x, y], [x, y],…]]
Return type:	list of list of floats

Example

In the following example all bounding boxes of the image annotation will be returned in list style:

>>> img_anno.anno_vec()
[[0.1 , 0.2 , 0.3 , 0.18],
 [0.25, 0.25, 0.2, 0.4]]

iter_annos(anno_type='bbox')[source]¶

Iterator for all related 2D annotations of this image.

Parameters:	anno_type (str) – Can be bbox’, ‘point’, ‘line’, ‘polygon’, ‘all’

Retruns:: iterator of TwoDAnno objects

Example

>>> for bb in img_anno.iter_annos('bbox'):
...     do_something(bb)

to_df()[source]¶

Tranform this ImageAnnotation and all related TwoDAnnotaitons into a pandas DataFrame.

Returns:

Column names are:: ’img_uid’, ‘img_timestamp’, ‘img_state’, ‘img_sim_class’, ‘img_frame_n’, ‘img_path’, ‘img_iteration’, ‘img_user_id’, ‘img_anno_time’, ‘img_lbl’, ‘img_lbl_id’, ‘img_user’, ‘img_is_junk’, ‘img_fs_name’, ‘anno_uid’, ‘anno_timestamp’, ‘anno_state’, ‘anno_dtype’, ‘anno_sim_class’, ‘anno_iteration’, ‘anno_user_id’, ‘anno_user’, ‘anno_confidence’, ‘anno_time’, ‘anno_lbl’, ‘anno_lbl_id’, ‘anno_style’, ‘anno_format’, ‘anno_comment’, ‘anno_data’

Return type: pandas.DataFrame

to_dict(style='flat')[source]¶

Transform this ImageAnno and all related TwoDAnnos into a dict.

Parameters:	style (str) – ‘flat’ or ‘hierarchical’. Return a dict in flat or nested style.
Returns:	In ‘flat’ style return a list of dicts with one dict per annotation. In ‘hierarchical’ style, return a nested dictionary.
Return type:	list of dict OR dict

Example

HowTo iterate through all TwoDAnnotations of this ImageAnno dictionary in flat style:

>>> for d in img_anno.to_dict():
...     print(d['img_path'], d['anno_lbl'], d['anno_dtype'])
path/to/img1.jpg [] None
path/to/img1.jpg ['Aeroplane'] bbox
path/to/img1.jpg ['Bicycle'] point

Possible keys in flat style:

>>> img_anno.to_dict()[0].keys()
dict_keys([
    'img_uid', 'img_timestamp', 'img_state', 'img_sim_class',
    'img_frame_n', 'img_path', 'img_iteration', 'img_user_id',
    'img_anno_time', 'img_lbl', 'img_lbl_id', 'img_user',
    'img_is_junk', 'img_fs_name', 'anno_uid', 'anno_timestamp',
    'anno_state', 'anno_dtype', 'anno_sim_class', 'anno_iteration',
    'anno_user_id', 'anno_user', 'anno_confidence', 'anno_time',
    'anno_lbl', 'anno_lbl_id', 'anno_style', 'anno_format',
    'anno_comment', 'anno_data'
])

HowTo iterate through all TwoDAnnotations of this ImageAnno dictionary in hierarchical style:

>>> h_dict = img_anno.to_dict(style='hierarchical')
>>> for d in h_dict['img_2d_annos']:
...     print(h_dict['img_path'], d['anno_lbl'], d['anno_dtype'])
path/to/img1.jpg [Aeroplane] bbox
path/to/img1.jpg [Bicycle] point

Possible keys in hierarchical style:

>>> h_dict = img_anno.to_dict(style='hierarchical')
>>> h_dict.keys()
dict_keys([
    'img_uid', 'img_timestamp', 'img_state', 'img_sim_class',
    'img_frame_n', 'img_path', 'img_iteration', 'img_user_id',
    'img_anno_time', 'img_lbl', 'img_lbl_id', 'img_user',
    'img_is_junk', 'img_fs_name', 'img_2d_annos'
])
>>> h_dict['img.twod_annos'][0].keys()
dict_keys([
    'anno_uid', 'anno_timestamp', 'anno_state', 'anno_dtype',
    'anno_sim_class', 'anno_iteration', 'anno_user_id',
    'anno_user', 'anno_confidence', 'anno_time', 'anno_lbl',
    'anno_lbl_id', 'anno_style', 'anno_format', 'anno_comment',
    'anno_data'
])

to_vec(columns='all')[source]¶

Transform this ImageAnnotation and all related TwoDAnnotations in list style.

Parameters: columns (str or list of str) – ‘all’ OR ‘img_uid’, ‘img_timestamp’, ‘img_state’, ‘img_sim_class’, ‘img_frame_n’, ‘img_path’, ‘img_iteration’, ‘img_user_id’, ‘img_anno_time’, ‘img_lbl’, ‘img_lbl_id’, ‘img_user’, ‘img_is_junk’, ‘img_fs_name’, ‘anno_uid’, ‘anno_timestamp’, ‘anno_state’, ‘anno_dtype’, ‘anno_sim_class’, ‘anno_iteration’, ‘anno_user_id’, ‘anno_user’, ‘anno_confidence’, ‘anno_time’, ‘anno_lbl’, ‘anno_lbl_id’, ‘anno_style’, ‘anno_format’, ‘anno_comment’, ‘anno_data’

Retruns:: list OR list of lists: Desired columns

Example

Return just a list of 2d anno labels:

>>> img_anno.to_vec('anno_lbl')
[['Aeroplane'], ['Bicycle']]

Return a list of lists:

>>> img_anno.to_vec(['img_path', 'anno_lbl'])
[
    ['path/to/img1.jpg', ['Aeroplane']],
    ['path/to/img1.jpg', ['Bicycle']]
]

TwoDAnno¶

class lost.db.model.TwoDAnno(anno_task_id=None, user_id=None, timestamp=None, state=None, track_id=None, sim_class=None, img_anno_id=None, timestamp_lock=None, iteration=0, data=None, dtype=None, confidence=None, anno_time=None, description=None, meta=None, is_example=False, meta_blob=None)[source]¶

A TwoDAnno represents a 2D annotation/ drawing for an image.

A TwoDAnno can be of type point, line, bbox or polygon.

idx¶

ID of this TwoDAnno in database

Type:	int

anno_task_id¶

ID of the anno_task this TwoDAnno belongs to.

Type:	int

timestamp¶

Timestamp created of TwoDAnno

Type:	DateTime

timestamp_lock¶

Timestamp locked in view

Type:	DateTime

state¶

can be unlocked, locked, locked_priority or labeled (see lost.db.state.Anno)

Type:	enum

track_id¶

The track id this TwoDAnno belongs to.

Type:	int

sim_class¶

The similarity class this anno belong to. It is used to cluster similar annos in MIA.

Type:	int

iteration¶

The iteration of a loop when this anno was created.

Type:	int

user_id¶

Id of the annotator.

Type:	int

img_anno_id¶

ID of ImageAnno this TwoDAnno is appended to

Type:	int

data¶

drawing data (for e.g. x,y, width, height) of anno - depends on dtype

Type:	Text

dtype¶

type of TwoDAnno (for e.g. bbox, polygon) (see lost.db.dtype.TwoDAnno)

Type:	int

labels¶

A list of Label objects related to the TwoDAnno.

Type:	list

confidence¶

Confidence of Annotation.

Type:	float

anno_time¶: Overall Annotation Time in ms.

description¶

Description for this annotation. Assigned by an annotator or algorithm.

Type:	str

meta¶

A field for meta information added by a script

Type:	str

is_example¶

Indicates wether this annotation is an example for the selected label.

Type:	bool

add_label(label_leaf_id)[source]¶

Add a label to this 2D annotation.

Parameters:	label_leaf_id (int) – Id of the label_leaf that should be added.

bbox¶

BBOX annotation in list style [x, y, w, h]

Example

>>> anno = TwoDAnno()
>>> anno.bbox = [0.1, 0.1, 0.2, 0.2]
>>> anno.bbox
[0.1, 0.1, 0.2, 0.2]

Type:	list

get_anno_serialization_format()[source]¶

Get annotation data in list style parquet serialization.

Returns:	For a POINT: [[ x, y ]] For a BBOX: [[ x, y, w, h ]] For a LINE and POLYGONS: [[x, y], [x, y],…]
Return type:	list of floats

Example

HowTo get a numpy array? In the following example a bounding box is returned:

>>> np.array(twod_anno.get_anno_vec())
array([0.1 , 0.2 , 0.3 , 0.18])

get_anno_vec()[source]¶

Get annotation data in list style.

Returns:	For a POINT: [x, y] For a BBOX: [x, y, w, h] For a LINE and POLYGONS: [[x, y], [x, y],…]
Return type:	list of floats

Example

HowTo get a numpy array? In the following example a bounding box is returned:

>>> np.array(twod_anno.get_anno_vec())
array([0.1 , 0.2 , 0.3 , 0.18])

line¶

LINE annotation in list style [[x, y], [x, y], …]

Example

>>> anno = TwoDAnno()
>>> anno.line = [[0.1, 0.1], [0.2, 0.2]]
>>> anno.line
[[0.1, 0.1], [0.2, 0.2]]

Type:	list of list

point¶

POINT annotation in list style [x, y]

Example

>>> anno = TwoDAnno()
>>> anno.point = [0.1, 0.1]
>>> anno.point
[0.1, 0.1]

Type:	list

polygon¶

polygon annotation in list style [[x, y], [x, y], …]

Example

>>> anno = TwoDAnno()
>>> anno.polygon = [[0.1, 0.1], [0.2, 0.1], [0.15, 0.2]]
>>> anno.polygon
[[0.1, 0.1], [0.2, 0.1], [0.15, 0.2]]

Type:	list of list

to_df()[source]¶

Transform this annotation into a pandas DataFrame

Returns:	A DataFrame where column names correspond to the keys of the dictionary returned from to_dict() method.
Return type:	pandas.DataFrame

Note

Column names are:: ‘anno_uid’, ‘anno_timestamp’, ‘anno_state’, ‘anno_dtype’, ‘anno_sim_class’, ‘anno_iteration’, ‘anno_user_id’, ‘anno_user’, ‘anno_confidence’, ‘anno_time’, ‘anno_lbl’, ‘anno_lbl_id’, ‘anno_style’, ‘anno_format’, ‘anno_comment’, ‘anno_data’

to_dict(style='flat')[source]¶

Transform this object into a dict.

Parameters:	style (str) – ‘flat’ or ‘hierarchical’ ‘flat’: Return a dictionray in table style ‘hierarchical’: Return a nested dictionary

Retruns:: dict: In flat or hierarchical style.

Example

Get a dict in flat style. Note that ‘anno.data’, ‘anno.lbl.idx’, ‘anno.lbl.name’ and ‘anno.lbl.external_id’ are json strings in contrast to the hierarchical style.

>>> bbox.to_dict(style='flat')
{
    'anno_uid': 1,
    'anno_timestamp': datetime.datetime(2022, 10, 27, 11, 27, 31),
    'anno_state': 4,
    'anno_dtype': 'point',
    'anno_sim_class': None,
    'anno_iteration': 0,
    'anno_user_id': 1,
    'anno_user': 'admin',
    'anno_confidence': None,
    'anno_time': 2.5548,
    'anno_lbl': ['Person'],
    'anno_lbl_id': [16],
    'anno_style': 'xy',
    'anno_format': 'rel',
    'anno_comment': None,
    'anno_data': [[0.5683337459767269, 0.3378842004739504]]}
}

to_vec(columns='all')[source]¶

Tansfrom this annotation in list style.

Parameters:	columns (list of str OR str) – Possible column names are: ‘all’ OR ‘anno_uid’, ‘anno_timestamp’, ‘anno_state’, ‘anno_dtype’, ‘anno_sim_class’, ‘anno_iteration’, ‘anno_user_id’, ‘anno_user’, ‘anno_confidence’, ‘anno_time’, ‘anno_lbl’, ‘anno_lbl_id’, ‘anno_style’, ‘anno_format’, ‘anno_comment’, ‘anno_data’
Returns:	A list of the desired columns.
Return type:	list of objects

Example

If you want to get only the annotation in list style e.g. [xc, yc, w, h] (if this TwoDAnnotation is a bbox).

>>> anno.to_vec('anno_data')
[0.1, 0.1, 0.2, 0.2]

LabelLeaf¶

class lost.db.model.LabelLeaf(idx=None, name=None, abbreviation=None, description=None, timestamp=None, external_id=None, label_tree_id=None, is_deleted=None, parent_leaf_id=None, is_root=None, group_id=None, color=None)[source]¶

A LabelLeaf

idx¶

ID in database.

Type:	int

name¶

Name of the LabelName.

Type:	str

abbreviation¶

Type:	str

description¶

Type:	str

timestamp¶

Type:	DateTime

external_id¶

Id of an external semantic label system (for e.g. synsetid of wordnet)

Type:	str

is_deleted¶

Type:	Boolean

is_root¶

Indicates if this leaf is the root of a tree.

Type:	Boolean

parent_leaf_id¶

Reference to parent LabelLeaf.

Type:	Integer

group_id¶

Group this Label Leaf belongs to

Type:	int

color¶

Color of the label in Hex format.

Type:	str

label_leafs¶

Type:	list of `LabelLeaf`

Note

group_id is None if this filesystem is available for all users!

to_df()[source]¶

Transform this LabelLeaf to a pandas DataFrame.

Returns:
Return type:	pd.DataFrame

to_dict()[source]¶

Transform this object to a dict.

Returns:
Return type:	dict

Label¶

class lost.db.model.Label(idx=None, dtype=None, label_leaf_id=None, img_anno_id=None, two_d_anno_id=None, annotator_id=None, timestamp_lock=None, timestamp=None, confidence=None, anno_time=None)[source]¶

Represants an Label that is related to an annoation.

idx¶

ID in database.

Type:	int

dtype¶

lost.db.dtype.Result type of this attribute.

Type:	enum

label_leaf_id¶: ID of related model.LabelLeaf.

img_anno_id¶

Type:	int

two_d_anno_id¶

Type:	int

timestamp¶

Type:	DateTime

timestamp_lock¶

Type:	DateTime

label_leaf¶

related model.LabelLeaf object.

Type:	model.LabelLeaf

annotator_id¶

GroupID of Annotator who has assigned this Label.

Type:	Integer

confidence¶

Confidence of Annotation.

Type:	float

anno_time¶

Time of annotaiton duration

Type:	float

logic.label¶

LabelTree¶

class lost.logic.label.LabelTree(dbm, root_id=None, root_leaf=None, name=None, logger=None, group_id=None)[source]¶

A class that represants a LabelTree.

Parameters:	dbm (`lost.db.access.DBMan`) – Database manager object. root_id (int) – label_leaf_id of the root Leaf. root_leaf (`lost.db.model.LabelLeaf`) – Root leaf of the tree. name (str) – Name of a label tree. logger (logger) – A logger. group_id (int) – Id of the group where the LabelTree belongs to.

create_child(parent_id, name, external_id=None)[source]¶

Create a new leaf in label tree.

Parameters:	parent_id (int) – Id of the parend leaf. name (str) – Name of the leaf e.g the class name. external_id (str) – Some id of an external label system.

Retruns:: lost.db.model.LabelLeaf: The the created child leaf.

create_root(name, external_id=None)[source]¶

Create the root of a label tree.

Parameters:	name (str) – Name of the root leaf. external_id (str) – Some id of an external label system.

Retruns:

lost.db.model.LabelLeaf or None:: The created root leaf or None if a root leaf with same name is already present in database.

delete_subtree(leaf)[source]¶

Recursive delete all leafs in subtree starting with leaf

Parameters:	leaf (`lost.db.model.LabelLeaf`) – Delete all childs of this leaf. The leaf itself stays.

delete_tree()[source]¶: Delete whole tree from system

get_child_vec(parent_id, columns='idx')[source]¶

Get a vector of child labels.

Parameters:	parent_id (int) – Id of the parent leaf. columns (str or list of str) – Can be any attribute of `lost.db.model.LabelLeaf` for example ‘idx’, ‘external_idx’, ‘name’ or a list of these e.g. [‘name’, ‘idx’]

Example

>>> label_tree.get_child_vec(1, columns='idx')
[2, 3, 4]

>>> label_tree.get_child_vec(1, columns=['idx', 'name'])
[
    [2, 'cow'],
    [3, 'horse'],
    [4, 'person']
]

Returns:
Return type:	list in the requested columns

import_df(df)[source]¶

Import LabelTree from DataFrame

Parameters:	df (pandas.DataFrame) – LabelTree in DataFrame style.

Retruns:

lost.db.model.LabelLeaf or None:: The created root leaf or None if a root leaf with same name is already present in database.

to_df()[source]¶

Transform this LabelTree to a pandas DataFrame.

Returns:	pandas.DataFrame

dtype¶

TwoDAnno¶

class lost.db.dtype.TwoDAnno[source]¶

Type of a TwoDAnno

BBOX¶

A BBox.

Type:	1

POLYGON¶

A Polygon.

Type:	2

POINT¶

A Point.

Type:	3

LINE¶

A Line.

Type:	4

CIRCLE¶

A Circle.

Type:	5

util methods¶

anno_helper¶

A module with helper methods to tranform annotations into different formats and to crop annotations from an image.

lost.pyapi.utils.anno_helper.calc_box_for_anno(annos, types, point_padding=0.05)[source]¶

Calculate a bouning box for an arbitrary 2DAnnotation.

Parameters:	annos (list) – List of annotations. types (list) – List of types. point_padding (float, optional) – In case of a point we need to add some padding to get a box.
Returns:	A list of bounding boxes in format [[xc,yc,w,h],…]
Return type:	list

lost.pyapi.utils.anno_helper.crop_boxes(annos, types, img, context=0.0, draw_annotations=False)[source]¶

Crop a bounding boxes for TwoDAnnos from image.

Parameters:	annos (list) – List of annotations. types (list) – List of types. img (numpy.array) – The image where boxes should be cropped from. context (float) – The context that should be added to the box. draw_annotations (bool) – If true, annotation will be painted inside the crop.
Returns:	A tuple that contains a list of image crops and a list of bboxes [[xc,yc,w,h],…]
Return type:	(list of numpy.array, list of list of float)

lost.pyapi.utils.anno_helper.divide_into_patches(img, x_splits=2, y_splits=2)[source]¶

Divide image into x_splits*y_splits patches.

Parameters:

img (array) – RGB image (skimage.io.imread).
x_splits (int) – Number of elements on x axis.
y_splits (int) – Number of elements on y axis.

Returns:

img_patches, box_coordinates: img batches and box coordinates of these patches in the image.

Return type:

list, list

Note

img_patches are in following order:: [[x0,y0], [x0,y1],…[x0,yn],…,[xn,y0], [xn, y1]…[xn,yn]]

lost.pyapi.utils.anno_helper.draw_annos(annos, types, img, color=(255, 0, 0), point_r=2)[source]¶

Draw annotations inside a image

Parameters:	annos (list) – List of annotations. types (list) – List of types. img (numpy.array) – The image to draw annotations in. color (tuple) – (R,G,B) color that is used for drawing.

Note

The given image will be directly edited!

Returns:	Image with drawn annotations
Return type:	numpy.array

lost.pyapi.utils.anno_helper.to_abs(annos, types, img_size)[source]¶

Convert relative annotation coordinates to absolute ones

Parameters:	annos (list of list) – types (list of str) – img_size (tuple) – (width, height) of the image in pixels.
Returns:	Annotations in absolute format.
Return type:	list of list

lost.pyapi.utils.anno_helper.trans_boxes_to(boxes, convert_to='minmax')[source]¶

Transform a box from standard lost format into a different format

Parameters:	boxes (list of list) – Boxes in standard lost format [[xc,yc,w,h],…] convert_to (str) – minmax -> [[xmim,ymin,xmax,ymax]…]
Returns:	Converted boxes.
Return type:	list of list

blacklist¶

A helper module to deal with blacklists.

class lost.pyapi.utils.blacklist.ImgBlacklist(my_script, name='img-blacklist.json', context='pipe')[source]¶

A class to deal with image blacklists.

Such blacklists are often used for annotation loops, in order to prevent annotating the same image multiple times.

my_script¶

The script instance that creates this blacklist.

Type:	`lost.pyapi.script.Script`

name¶

The name of the blacklist file.

Type:	str

context¶

Options: instance, pipe, static

Type:	str

Example

Add images to blacklist.

>>> blacklist = ImgBlacklist(self, name='blacklist.json')
>>> blacklist.add(['path/to/img0.jpg'])
>>> balcklist.save()

Load a blacklist and check if a certain image is already in list.

>>> blacklist = ImgBlacklist(self, name='blacklist.json')
>>> blacklist.contains('path/to/img0.jpg')
True
>>> blacklist.contains('path/to/img1.jpg')
False

Get list of images that are not part of the blacklist

>>> blacklist.get_whitelist(['path/to/img0.jpg', 'path/to/img1.jpg', 'path/to/img2.jpg'])
['path/to/img1.jpg', 'path/to/img2.jpg']

Add images to the blacklist

>>> blacklist.add(['path/to/img1.jpg', 'path/to/img2.jpg'])

add(imgs)[source]¶

Add a list of images to blacklist.

Parameters:	imgs (list) – A list of image identifiers that should be added to the blacklist.

contains(img)[source]¶

Check if blacklist contains a spcific image

Parameters:	img (str) – The image identifier
Returns:	True if img in blacklist, False if not.
Return type:	bool

delete_blacklist()[source]¶: Remove blacklist from filesystem

get_whitelist(img_list, n='all')[source]¶

Get a list of images that are not part of the blacklist.

Parameters:	img_list (list of str) – A list of images where should be checked if they are in the blacklist n ('all' or 'int') – The maximum number of images that should be returned.
Returns:	A list of images that are not in the blacklist.
Return type:	list of str

remove_item(item)[source]¶

Remove item from blacklist

Parameters:	item (str) – The item/ image to remove from blacklist.

Developing Pipelines¶

LOST Ecosystem¶

All about Pipelines¶

Pipeline Projects¶

Pipeline Project Examples¶

Repository Structure¶

Importing a Pipeline Project into LOST¶

Updating a LOST Pipeline¶

Namespacing¶

Pipeline Definition Files¶

An Example¶

Possible Pipeline Elements¶

All about Scripts¶

pyapi¶

inout¶

ScriptOutput¶

Output¶

Input¶

pipeline¶

PipeInfo¶

pipe_elements¶

Datasource¶

AnnoTask¶

MIATask¶

SIATask¶

DataExport¶

VisualOutput¶

Loop¶

model¶

ImageAnno¶

TwoDAnno¶

LabelLeaf¶

Label¶

logic.label¶

LabelTree¶

dtype¶

TwoDAnno¶

util methods¶

anno_helper¶

blacklist¶

vis¶

Importing / Updating Pipelines¶

Import via GUI¶

LOST cli¶