Layout
Tool Components and Features
Header
The header consists of 3 primary things - Focus mode to view 3D cuboids from 3 angles, control point cloud orientation and visibility and navigation controls.
Rotate Direction
The first button beside the tool name re-orients the direction of the selected cuboid annotation by 90
Shortcut to Rotate cuboid by 90 degrees : R
Focus Mode
On enabling Focus mode, the selected object is focused and the header displays more options. The user sees the cuboid in a 2 dimensional view from all sides.
Shortcuts for Jumping side:
Jump to Front: Shift+A
Jump to Right: Shift+S
Jump to Top: Shift_D
Control Point Cloud View
Switching between the below two modes allows users to mark objects accurately.
2D Mode
Allows users to view the point cloud from the top view. User creates a rectangle in this mode.
3D Mode
Allows users to view the point cloud in 3D. Creating a cuboid in this mode allows users to drop a cuboid in a 3D angle.
Sensor
The ego vehicle has multiple lidar sensors installed on it. Because the sensors are not time-synced, a staggering effect of the objects around the ego vehicle can be observed in the point cloud (refer to image below). The user can use the sensor filter and select the sensor points they would like to consider.
Radar
Within the Sensor feature, the Radar data can be togglable
AutoQA
Users can AutoQA their tasks by completing the work expected from them on a frame and then clicking on the shield. This shield will execute a predefined set of logical validations and return with a report.
The report will present the following information of the annotation which breaks the validation:
Class and ID (for e.g. Human_1, Vehicle_1, Vegetation, etc.)
Object Type (for e.g. cuboid, 2D bbox, polyline, etc.)
Type (warning/mandatory)
Error value (the attribute value, dimension, position, etc. that didn't meet the criteria)
Description (the description provided to the validation)
The validations are configurable to suit the workflow.
Isolate
Users can crop out an area in the point cloud to better focus on the object. Once cropped out, the user can create an annotation. This is done by toggling the ‘Isolate’ option and creating a polygon around the area. Once done, select a preferred drawable tool and annotate the points.
Outlier
Users have the option to view the unmarked points by selecting the toggle present in the header.
Navigation mode
Users can navigate through the point cloud using Keyboard and mouse interactions. Both the modes allow different navigation motions.
Helicopter Mode (Shortcuts):
WASD : Move forward/ Left/ Back/ Right
QE : Fly Up/ Down
Arrow keys to look around
Mouse wheel : Zoom control
Orbital Mode (Shortcuts):
WASD : Orbit around Forward/ Right/ Backwards/ Left
Arrow keys to slide Top/ Left/ Down Right
Mouse wheel : Zoom control
Fit to screen
This fits the entire point cloud within the browser screen
Jump to Origin
The coordinates at which the user is positioned by default when the task loads is the Point of Origin. On clicking this, the user is navigated back to this point.
Toggle full screen
Hides browser bar and makes tool fullscreen.
Status
After selecting a cuboid, the Status button on the top right corner is enabled. On viewing the Status, we can see the following information real-time
dimension (length, width, height) in meters,
rotation values (yaw, pitch, roll) in degrees and
position of the centroid in X, Y, Z coordinate system
Issues
In the first round, when the annotator begins the annotation and/or when a reviewer reviews the annotation data, the Issues button is disabled.
When the same batch moves to Audit stage and gets reviewed by the auditor, the Issues panel is enabled. The Auditor raises feedback in the form of issue cards as and when they spot a mistake. This batch, when finished on the Audit node, can be reopened on the Annotator/Reviewer nodes with the issue cards visible. This aids the annotator to review the issues marked and fix the quality of the sequence and resolve the issues.
Help
Users can find the Shortcuts used in the tool and the Batch Information under this menu.
Shortcuts
Opens a panel to display Tool Shortcuts
Batch Information
Information about the batch is provided under this option
Footer
Time tracker
Tracks how much time is spent on a task and displays it real-time.
Timeline
The timeline displays all the frames within a sequence
User can move from one frame to another by dragging and dropping the blue marker on the timeline
The user can drag and drop the timeline to move to a particular frame
When no cuboid annotation is selected, the timeline displays the current frame
When a cuboid annotation is selected,the timeline displays the occurrence of the annotation across the frames by the class color
On expanding the timeline, one can see the values of all the attributes of the selected annotation on independent rows.
If the attribute has multiple values, it can be seen on hover.
Time Varying and Constant Attribute
Constant Attributes
The value of this attribute remains constant across the frame. Hence if change on any frame, this new value is reflected across the frames (indicated in the expanded timeline).
Time Varying Attributes
The value of this attribute can vary across the frames. The same is reflected in the expanded timeline view with color changes ensuring correct labeling for the annotated object throughout the entire sequence.
Sensory Attributes
This attribute allows annotations to capture details that are unique to individual sensors. This attribute is a sub-type of Time Varying attribute
Object Attributes and Interpolation
Editing a class level attribute creates a new keyframe, and attribute values are copied between keyframes. This means that if the user edits a class level attribute, its value is copied until the next existing keyframe.
If the user wants to change an attribute of an object with existing keyframes, make sure to change the value on all necessary keyframes.
Go To Frame (Task) Navigator
Navigate between tasks from here. If the Skip Frames option is not enabled in the Taskflow, the tool compels the annotator to first visit all frames chronologically and then allow them to jump between frames.
Menu
Tool Settings
Point Cloud visibility can be configured here
Timer Range
Each point captured by the Lidar sensor has a timestamp. This timestamp is the time difference between the reference time of the frame and the time of an individual point. In the below example, '0' is the frame's reference time. Using the Timer Ranger filter, the user can select any range between -49 up to +49 and filter the points in between. Keeping the controls of the slider at the end will show the entire point cloud.
Point Budget
Whenever the tool loads the frame, this slider decides the number of points to load. By default, the value is set to 1 million. If the current frame contains more than a million points, then the tool loads at most a million points in the area you are viewing. Zooming in loads the other points that were not loaded.
Also, with increasing the limit of points, latency is to be expected.
Enable Eye Dome Lighting
EDL is a lighting model that accentuates the shapes of objects within a point cloud by grouping objects that are close together and shading their outlines, which enhances depth perception.
Class Point Brightness
The points that have been classified appear in their class color. With the help of this slider user can reduce the brightness of those colored points.
Background
Black, White, Gradient
Change the background color to make use of the different light situations.
Point size modifier:
Increases/decreases the size of the points on the screen.
Point Colors
View the points in different color schemes
PlainColor
The points are displayed in a single color which the user can choose from the palette.
Gradient
The points are displayed based on their height, with cooler colors indicating lower positions on the Z-axis. With the help of Gradient Range: you can spread out the colors.
RGBA
When the point cloud data contain an RGB field, this information is used to set the color of the points.
Intensity
When the point cloud files contain an intensity field, this information is used to set the color of the points.
Speed Controls
The speed of a mouse scroll can be calibrated. The higher the value the further distance is traveled in one scroll.
Region of Interest
User can define a region of interest from the ego vehicle, by defining the radius from the ego. Multiple such concentric circle at different radius’ can be created.
Task Attributes
Task related attributes are listed in the left rail, once the user clicks on this sub-menu. These are defined while creating a recipe and tagging an attribute at a task level.
Batch Attributes
Task related attributes are listed in the left rail, once the user clicks on this sub-menu. These are defined while creating a recipe and tagging an attribute at a batch level.
Add Comment
Annotators can add any doubts or comments in this section. The reviewer is notified about the same when the task is assigned to them in the next node.
Check Task
If there are any current task/frame related constraints configured during project setup, they appear here. Users can verify if they have completed the task and move to the next task.
Left Panel
Class List
The Left panel primarily consists of all the Classes that need to be marked in the task. These Classes are defined in the class step of the ‘Create a Recipe’ process.
Object Visibility Controls
Filled Boxes : Fills the cuboid with color
Object Names
Displays name of all annotations on both the point cloud and reference image
Right Side
Ground/Ceiling Level
On the bottom right of the point cloud is a two-way slider that allows the user to hide points from the ceiling and ground as per requirement. On hiding unnecessary points, users can accurately mark the visible points.
Shortcut to change Levels:
Ground Level (increase/ decrease) : G + Up Arrow/ Down Arrow
Ceiling Level (increase/ decrease) : C + Up Arrow/ Down Arrow
Show all points
This filter allows the user to view only the ground points, or everything above ground points. This segregation of a point being ground or non-ground occurring during the preprocess stage. The data is run through an ML algorithm and then loaded into the tool.
In case ML is not needed, the ground segregation is done through the algorithm patchwork++.
Last updated