Labeling
This page provides basic description of marker labels and instructions on labeling workflow in Motive.
Labeling: Basic Concept
Marker Label
Marker labels are basically software name tags that are assigned to trajectories of reconstructed 3D markers so that they can be referenced for tracking individual markers, Rigid Bodies, or Skeletons. Motive identifies marker trajectories using the assigned labels. Labeled trajectories can be exported individually, or combined together to compute positions and orientations of the tracked objects. In most applications, all of the target 3D markers will need to be labeled in Motive. There are two methods for labeling markers in Motive: auto-labeling and manual labeling, and both labeling methods will be covered in this page.
Monitoring Labels
Labeled or unlabeled trajectories can be identified and resolved from the following places in Motive:
3D Perspective Viewport: From the 3D viewport, check the Marker Labels in the visual aids option to view marker labels for selected markers.
Labels pane: The Labels pane lists out all of the marker labels and corresponding percentage gap for each label. The color of the label also indicates whether if the label is present or missing at the current frame.
Graph View pane: For frames where the selected label is not assigned to any markers, the timeline scrubber gets highlighted in red. Also, the tracks view of this pane provides a list of labels and their continuity in a captured Take.
Labeling Methods
There are two approaches to labeling markers in Motive:
Auto-label pipeline: Automatically label sets of Rigid Body markers and Skeleton markers using calibrated asset definitions.
Manual Label: Manually label individual markers using the Labels pane.
For tracking Rigid Bodies and Skeletons, Motive can use the asset definitions to automatically label associated markers both in real-time and post-processing. The auto-labeler uses references assets that are enabled, or assets that are checked in the Assets pane, to search for a set of markers that matches with the definition and assign pre-defined labels throughout the capture.
There are times, however, when it is necessary to manually label a section or all of a trajectory, either because the markers of a Rigid Body or a Skeleton were misidentified (or unidentified) during capture or because individual markers need to be labeled without using any tracking assets. In these cases, the Labels pane in Motive is used to perform manual labeling of individual trajectories. Manual labeling workflow is supported only in post-processing of capture when a Take file (TAK) has been loaded with 3D data as its playback type. In case of 2D data only capture, the Take must be Reconstructed first in order to assign, or edit, the marker labels in its 3D data. This manual labeling process, along with 3D data editing is typically referred to as post processing of mocap data.
Auto-label
Rigid body and Skeleton asset definitions contain information of marker placements on corresponding assets. This is recorded when the assets are first created, and the auto-labeler in Motive uses them to label a set of reconstructed 3D trajectories that resemble marker arrangements of active assets. Once all of the markers on active assets are successfully labeled, corresponding Rigid Bodies and Skeletons get tracked in the 3D viewport.
The auto-labeler runs in real-time during Live mode and the marker labels get saved onto the recorded TAKs. Running the auto-labeler again in post-processing will basically attempt to label the Rigid Body and Skeleton markers again from the 3D data.
Auto-labeling Steps
From Data pane
Select Takes from the Data pane
Right-click to bring up the context menu
Click reconstruct and auto-label' to process selected Takes. The this pipeline will create a new set of 3D data and auto-label the markers from it.
This will label all the markers that matches the corresponding asset definition.
Auto-labeler Settings
The settings for the auto-labeling engine are defined in the Auto-labeler section of the Reconstruction pane. The auto-labeler parameters can be modified during post-processing pipelines, and they can be optimized for stable labeling of markers throughout the Take.
Note: Be careful when reconstructing a Take again either by Reconstruct or Reconstruct and Auto-label, because it will overwrite the 3D data and any post-processing edits on trajectories and marker labels will be discarded. Also, for Takes involving Skeleton assets, the recorded Skeleton marker labels, which were intact during the live capture, may be discarded, and reconstructed markers may not be auto-labeled again if the Skeletons are never in well-trackable poses throughout the captured Take. This is another reason why you want to start a capture with a calibration pose (e.g. T-pose).
Marker Sets
The Marker Set is a type of assets in Motive. It is the most fundamental method of grouping related markers, and this can be used to manually label individual markers in post-processing of captured data using the Labeling pane. Note that Marker Sets are used for manual labeling only. For automatic labeling during live mode, a Rigid Body asset or a Skeleton asset is necessary.
Since creating rigid bodies, or skeletons, groups the markers in each set and automatically labels them, Marker Sets are not commonly used in the processing workflow. However, they are still useful for marker-specific tracking applications or when the marker labeling is done in pipelines other than auto-labeling. Also, marker sets are useful when organizing and reassigning the labels.
Create MarkerSet
Adding Labels
Labels pane
The Labels pane is used to assign, remove, and edit marker labels in the 3D data. The Tracks View under the Graph View pane can be used in conjunction with the Labels pane to monitor which markers and gaps are associated. The Labels pane is also used to examine the number of occluded gaps in each label, and it can be used along with the Editing Tools for complete post-processing.
For a given frame, all labels are color-coded. For each frame of 3D data, assigned marker labels are shown in white, labels without reconstructions are shown in red, and unlabeled reconstructions are shown in orange; similar to how they are presented in the 3D View.
See the Labels pane page for detailed explanation on each option.
QuickLabel Mode
The QuickLabel mode allows you to tag labels with single-clicks in the view pane, and it is a handy way to reassign or modify marker labels throughout the capture. When the QuickLabel mode is toggled, the mouse cursor switches to a finger icon with the selected label name attached next to it. Also, when the display label option is enabled in the perspective view, all of assigned marker labels will be displayed next to each marker in the 3D viewport, as shown in the image below. Select the marker set you wish to label, and tag the appropriate labels to each marker throughout the capture.
When assigning labels using the Quick Label Mode, the labeling scope is configured from the labeling range settings. You can restrict the labeling operation to apply from the current frame backward, current frame forward, or both depending on the trajectory. You may also restrict labeling operations to apply the selected label to all frames in the Take, to a selected frame range, or to a trajectory 'fragment' enclosed by gaps or spikes. The fragment/spike setting is used by default and this best identifies mislabeled frame ranges and assigns marker labels. See the Labels pane page for details on each feature.
Labeling using the QuickLabel Mode
Under the drop-down menu in the Labels pane, select an asset you wish to label.
All of the involved markers will be displayed under the columns.
From the label list, select unlabeled or mislabeled markers.
Inspect the behavior of the selected trajectory and decide whether you want to apply the selected label to frames forward or frames backward or both. This option can be selected from labeling range settings on the Labels pane.
Hiding Marker Labels
If the marker labels are set to visible in the 3D viewport, Motive will show all of the marker labels when entering the QuickLabel mode. To hide all of the marker labels from showing up in the viewport, you can click on the visual aids option in the perspective view, and uncheck marker labels.
General Labeling Steps
The following section provides the general labeling steps in Motive. Note that the labeling workflow is flexible and alternative approaches to the steps listed in this section could also be used. Utilize the auto-labeling pipelines in combination with the Labels pane to best reconstruct and label the 3D data of your capture.
Labeling Tips
Use the Graph View pane to monitor occlusion gaps and labeling errors as you post-process capture Takes
When using the Labeling pane, choose the most appropriate labeling setting (all, selected, spike, or fragment) to efficiently label selected trajectories. See more from the Labeling pane page.
Hotkeys can increase the speed of the workflow. Use Z and Shift+Z hotkeys to quickly find gaps in the selected trajectory.
When working with skeleton assets, label the hip segment first. The hip segment is the main parent segment, top of the segments hierarchy, where all other child segments are associated to. Manually assigning hip markers sometimes help the auto-labeler to label the entire asset.
For skeleton assets, the Show Tracking Errors property can be utilized to display tracking errors on skeleton segments.
Using Combined Reconstruction and Auto-label Pipeline
Step 1. In the Data pane, Reconstruct and auto-label the take with all of the desired assets enabled.
Step 2. In the Graph View pane, examine the trajectories and navigate to the frame where labeling errors are frequent.
Step 3. Open the Labels pane.
Step 4. Select an asset that you wish to label.
Step 5. From the label columns, Click on a marker label that you wish to re-assign.
Step 6. Inspect behavior of a selected trajectory and its labeling errors and set the appropriate labeling settings (allowable gap size, maximum spike and applied frame ranges).
Step 7. Switch to the QuickLabel mode (Hotkey: D).
Step 8. On the Perspective View, assign the labels onto the corresponding marker reconstructions by clicking on them.
Step 9. When all markers have been labeled, switch back to the Select Mode.
Using Standalone Reconstruction Pipeline and Auto-label Pipeline Separately
Step 1. Start with 2D data of a captured Take with model assets (Skeletons and Rigid Bodies).
Step 2. Reconstruct and Auto-Label, or just Reconstruct, the Take with all of the desired assets enabled under the Assets pane. If you use reconstruct only, you can skip step 3 and 5 for the first iteration.
Step 3. Examine the reconstructed 3D data, and inspect the frame range where markers are mislabeled.
Step 4. Using the Labels pane, manually fix/assign marker labels, paying attention to your label settings (direction, max gap, max spike, selected duration).
Step 5. Unlabel all trajectories you want to re-auto-label.
Step 6. Auto-Label the Take again. Only the unlabeled markers will get re-labeled, and all existing labels will be kept the same.
Step 7. Re-examine the marker labels. If some of the labels are still not assigned correctly from any of the frames, repeat the steps 3-6 until complete.
Labeling Error Fix
The general process for resolving labeling error is:
Identify the trajectory with the labeling error.
Determine if the error is a swap, an occlusion, or unlabeled.
Resolve the error with the correct tool.
Swap: Use the Swap Fix tool ( Edit Tools ) or just re-assign each label ( Labels panel ).
When manually labeling markers to fix swaps, set appropriate settings for the labeling direction, max spike, and selected range settings.
Occlusion: Use the Gap Fill tool ( Edit Tools ).
Unlabeled: Manually label an unlabeled trajectory with the correct label ( Labels panel ).
For more data editing options, read through the Data Editing page.
Sample Scenarios and Tutorial Videos
Following tutorials use Motive 1.10. On Motive 2.0., Data pane and Assets pane is used instead of the Project pane.
Sample Scenario 1 :: All Good
When recorded 3D data have been labeled properly and entirely throughout the Take, you will not need to edit marker labels. If you don't have 3D data recorded, you can reconstruct and auto-label the Take to obtain 3D data and label all of the skeleton and rigid body markers. If all of the markers are well reconstructed and there are no significant occlusions, auto-labeled 3D data may be acceptable right away. In this case, you can proceed without post-processing of marker labels.
Recorded 3D data has no gaps in the labels, or the Reconstruct and Auto-label works perfectly the first time without additional post-processing.
Examine the Take(s). Check the Labeling pane, or the tracks view, to make sure no occlusion exists within the capture, and all markers are consistently labeled.
Done.
Sample Scenario 2 :: Labeling errors in the middle of a Take
When skeleton markers are mislabeled only within specific frame ranges of a Take, you will have to manually re-label the markers. This may occur when a subject performs dynamic movements or come into contact with another object during the recorded Take. After correcting the mislabeled markers, you can also use the auto-labeler to assign remaining missing labels.
Start with recorded 3D data or Reconstruct and auto-label the Take to obtain newly labeled 3D data.
Inspect the Take to pick out the frame ranges with bad tracking.
If markers are mislabeled during majority of the capture, unlabel all markers from the entire capture by right-clicking on the Take in Data Management pane and click Delete Marker Labels. You can do this on selected frame ranges as well.
Scrub the timeline to a frame just before the bad tracking frame range.
Using the Labeling pane, manually label the skeleton. Depending on the severity of the mislabels, you can either label the entire skeleton or just the key segments starting from the hip.
Scrub the timeline to a frame after the bad tracking frame range.
Manually label the same skeleton.
Auto-label the Take.
Check the frames again and correct any remaining mislabels using the Labeling pane.
Sample Scenario 3 :: Skeletons never acquire perfectly throughout a Take
For Take(s) where skeletons are never perfectly tracked and the markers are consistently mislabeled, you will need to manually assign the correct labels for the skeleton asset(s). Situations like this could happen when the skeleton(s) are never in an easily trackable pose throughout the Take (e.g. captures where the actors are rolling on the ground). It is usually recommended that all skeleton ‘’Takes’’’ start and end with T-pose in order to easily distinguish the skeleton markers.
This also helps the skeleton solver to correctly auto-label the associated markers; however, in some cases, only specific section of a Take needs be trimmed out, or including the calibration poses might not be possible. Manually assigning labels can help the auto-labeler to correctly label markers and have skeletons acquire properly in a Take.
You will get best results if you manually label the entire skeleton, but doing so can be time-consuming. You can also label only the mislabeled segment or the key segment (hip bone) and run the auto-labeler to see if it correctly assigns the labels with the small help.
Start with recorded 3D data or Reconstruct the Take.
At a certain point of the Take (usually at a frame where you can best identify the pose of the skeleton), use the Labeling pane to manually assign the marker labels for skeletons that are not labeling correctly. Depending on the severity of the mislabels, you can either label the entire skeleton or only the key segments starting from the hip.
After manually assigning the labels, auto-label the Take. Make sure the corresponding assets are enabled in the Assets pane.
Check to see if all markers are correctly assigned throughout the take. If not, re-label or unlabel, any mislabeled markers and run auto-label again if needed.
Sample Scenario 4 :: Mislabeling/unlabeling of a marker due to occlusions
Marker occlusions can be critical to the auto-labeling process. After having a gap for multiple frames, occluded markers can be unlabeled entirely, or nearby reconstructions can be mistakenly recognized as the occluded marker and result in labeling swaps or mislabels. Skeleton and rigid body asset definitions may accommodate labeling for such occlusions, but in some cases, labeling errors may persist throughout the Take. The following steps can be used to re-assign the labels in this case.
If tracked markers are relatively stationary during the occluded frames, you may want to increase the Maximum Marker Label Gap value under the Auto-Labeler settings in the Reconstruction pane to allow the occluded marker to maintain its label after auto-labeling the Take. However, note that adjusting this setting will not be useful if the marker is moving dynamically beyond the Prediction Radius (mm) settings during occlusion.
Start with recorded 3D data or Reconstruct and auto-label the Take
Examine through the Take, and go to a frame where markers are mislabeled right after an occlusion.
Using the Quick Label Mode, correct the labeling errors.
Move onto next occluded frames. When the marker reappears, correct the labels.
After correcting the labels, Auto-label the Take again.
Use the Fill Gaps tool in the Editing tools to interpolate the occluded trajectories.
Last updated