Create Tasks
๐The first step to construct wholesome collaborative annotaion workflow is to create a task flow. Here is the detailed guide for you to start creating tasks.
Getting Started
Team Admin/Owner can click Team Task Management
tab under the Task module on BasicAI - > click Create a Task
tab to start creating a new task. The platform supports two creation methods: Create a Task by Yourself or Create a Task by Templates.
Reminder
Only Team Admin/Owner is able to create a task. Learn more details on Role & Permissions page.
How do I create a task flow?Please find the detailed guide below.
Hand-on-Hand Guide:
Create a Task by Yourself:
Step 2: Upload Data from Dataset
Step 3: Configure Ontology&Instructions
Step 4: Set up QA Rules (Optional)
Create a Task by Templates:
Create a Task by Yourself
The overall process of creating a task includes 5 steps: Task Basics, Upload Data, Ontology & Instructions, QA Rules and Workflow. Please refer to the following flow chart for more details.
Task&Data Type Elaborations:
BasicAI platform currently supports the creation of task types including LiDAR Fusion, Image, Audio & Video, and Text data types. The scene types and annotation types supported for publishing different data types are:
LiDAR Fusion: Supports publishing single data, scenes, 4D BEV, and the annotation type can be selected as 'instance' or 'segmentation'
Image: Supports publishing single data and scenes, and the annotation type can be selected as 'instance' or 'segmentation'
Audio & Video: Only supports publishing single data, and annotated in the type of 'instance'
Text: Only supports publishing single data, and the annotation type can be selected as 'instance' or 'relation'
Step 1: Fill in Task Basics
- Fill in Task Name and select
Create a Task by Yourself
- Select Data Type:
LiDAR Fusion
,Image
,Audio & Video
, orText
- To bind the task to a project, select a project from the
Link to a Project
drop-down box - Select Scenario Type: Scenario Type defines how to organize data in tasks, including
Single Data
,Scenes
,4D BEV
(only for LiDAR Fusion).Single Data
is designed for detection tasks only, whileScene
supports tracking tasks. - Select Annotation Type: Annotation Type determines how you would like to annotate your data, including
Instance
(annotation tools have bounding boxes, cuboids, polygons, etc.),Segmentation
, andRelations
(only for Text).
Step 2: Upload Data
You can either select an already existing dataset or upload a new one. To upload a new dataset, please find more details on the Upload Dataset page.
Please note: if you cannot find the data, please check if the data has been grouped into a scene in the dataset!
- Click on the previously-configured Dataset to upload data. The data type needs to be consistent.
- Click
+ Add
and select data from the dataset; clickSelect All
in the upper left corner to select all the data. After selecting the data, clickConfirm
to complete uploading.
Re-Group Scene Function
If you choose 'Scene' as the data type when creating a new task, we have a function called Re-Group Scene. After uploading a scene form a dataset, if the amount of data from the scene is too large, you can click
Regroup the Scene for Workers
and fill in thenumber of data as a Scene
input box to divide the large scene into smaller scenes, which is more convenient for workers to perform the task.
Step 3: Configure Ontology & Instructions
- Select Class & Classification from the configured ontology, or click
Edit
to jump to the Dataset module for ontology set-up. - Slide down the page, and if necessary, you can upload the operation instructions in PDF format for workers' reference.
Step 4: Set up QA Rules (Optional)
If you need to use automated QA functions for all the data in the task, you can enable the Rules
button and select the QA rules from the rule list. You can click Configure
on the right side of each rule to configure it, and click View Instruction
to view the rule details and descriptions. If you do not need to configure system QA, you can directly click Next
to enter the next page.
Step 5: Configure Workflow
Workflow Settings
On the
Workflow
page, the following options are provided for setting up:Choose task type: single-task or multi-task type,
Set task life stages: add review stages except annotation
Assign admins and workers: choose annotators, reviewers, inspectors and task admins
Select comment tags : choose comment tags from
Commments Pool
Complete Task & Annotate settings
1๏ธโฃ Task Distribution
Choose One Task
or Multiple Tasks
, depending on the amount of data to be annotated. If the amount of data is too large, you can choose to create Multiple Tasks.
- Select
One Task
: proceed with the subsequent setting options directly. - Select
Multiple Tasks
: you need to configure the amount of data for each task and the task naming rules, then click theRegenerate Tasks
button to generate the task details table. - Click the
Config
button in the task sheet to enter the subsequent configuration pages for each task.
2๏ธโฃ Workflow Setting
Set the life cycle to be experienced for tasks, assign workers to each stage, or share task jobs with external partners.
- Choose team members as annotators to do annotation work.
- Click
Add Stage
to add a review stage, and assign reviewers. Up to 5 review stages can be added. - Set up
Claim Size
(the number of data that each worker can claim at a time) andMax Claim Duration
(the maximum number of minutes allowed for finishing a claim of data) on each job card.
3๏ธโฃ Assign Admins and Inspector
Assign Task Admin and Inspector to the task. Only the inspector has permission to accept the task.
4๏ธโฃ Comments
Select comment tags that can be displayed on the tool page, such as Missed
,Wrong Object
,Wrong Label
,Not Fit
, etc.
5๏ธโฃ Task Settings
When to Start
controls when the task starts after it is published.Automatically
: Start immediately after the task is published.All Stages Ready
: Ifshare the task with partners
is selected in the workflow, the task will only start after all partners accept it.Wait for My Confirmation
: Admins need to start the task manually.
Data Accuracy
: Set the accuracy threshold of data to determine whether the data is accurate. The default threshold is 95%.- Choose whether to allow workers to view submitted data.
- Choose whether to allow group results to be included when calculating accuracy and miss rate.
6๏ธโฃ Annotation Settings
- Choose whether results must have ontology classes or classifications. If yes, choose
All Results
,Non-Group Results
, orGroup Results Only
. - Choose whether the result can be allowed to exceed the image area.
- Choose whether submission can be allowed if the result exceeds the Class limit.
- Enable the
Measure Tools
button to configure measure tools if you need to configure auxiliary lines, auxiliary tools, or dividing lines.
Step 6: Publish
After configuring all the above steps, click Publish
to complete task publishing. After publishing, all tasks will be displayed underย Team Task Management
.
Create a Task by Templates
Task Creation Templates
BasicAI provides users with several templates for quickly creating tasks for two types of data: 3D LiDAR and Image in the autonomous driving industry. The templates use prepared datasets and ontology settings, which can help users create tasks in a faster and more efficient way.
The currently available templates are shown in the table below. After clicking on the template card, detailed explanations of the template will appear on the right side of the page.
Data Type | Task Creation Templates | Remarks |
---|---|---|
3D LiDAR | Quick Start | 3D LiDAR task creation quick start |
3D LiDAR | PandaSet Detection Task | Suitable for creating PandaSet 2D/3D Object Detection tasks |
3D LiDAR | PandaSet Tracking Task | Based on 50 continuous frames from PandaSet and 4 real-world classes |
3D LiDAR | PandaSet DebugTask | Users can debug model results and then make comparisons in the dataset |
Image | CoCo Dataset Object Detection | Based on CoCo Dataset with 10 ontology classes without attributes |
Image | PandaSet Classification Task | Suitable for creating PandaSet image classification tasks |
After selecting a template, click
Next
to enter the subsequent steps sinceUpload data
, which are the same as explained in the Create a Task by Yourself section above, but the dataset and ontology configurations are provided by the selected template.
Next
Go to the Work with Tasks page to learn how to start annotation, review and acceptance, etc.
For Task Admin (Project Manager), please find more information on Manage Tasks page about how to manage tasks.
Updated 3 days ago