birkeh
/
o3de
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
11,665 Commits
3 Branches
3 Tags
148 MiB
development
monroegm-disable-blank-issue-2
main
2111.2
2111.1
2107.1
Branches Tags
${ item.name }
Create tag ${ searchTerm }
Create branch ${ searchTerm }
from '67fcf42268'
${ noResults }
Commit Graph

2 Commits (67fcf422685c36fbe5d3bbbb0a12640bf5576085)

Author SHA1 Message Date
Benjamin Jillich 3eb661c4bc Motion Matching: Moved discard event check into consolidated IsFrameDiscarded() method 
* Moved the extract events call and the check for a discard frame event to the IsFrameDiscarded() function.
* Added class description for the frame database and ported the comments to //! and //<

Signed-off-by: Benjamin Jillich <jillich@amazon.com>
 4 years ago
Benjamin Jillich 2068c225d1
Motion Matching (#7232) 
commit 0c873fc67f0250ec7155e1999c34930b23c7647f
Author: Benjamin Jillich <jillich@amazon.com>
Date:   Wed Jan 26 10:05:07 2022 +0100

    Motion Matching: Automated tests for the feature matrix and feature schema (#38)

    * Set up base motion matching test fixture
    * Automated tests for feature matrix
    * Automated tests for feature schema

    Signed-off-by: Benjamin Jillich <jillich@amazon.com>

commit cbd4f124481faf4c8bf0b98ae4602140f231e2fb
Author: Benjamin Jillich <jillich@amazon.com>
Date:   Tue Jan 25 09:59:32 2022 +0100

    Motion Matching: User adjustable algorithm via UI and new residual calculation option (#37)

    * Added the feature properties to the edit context. As all the preparation work has been done, this finalizes the adjustable algorithm UI work.
    * Added a new residual calculation type for calculating the differences between the input query calues and the features extracted from the motion database. Either the differences are just used as an absolute value or they are squared. "Use 'Squared' in case minimal differences should be ignored and larger differences should overweight others. Use 'Absolute' for linear differences and don't want the mentioned effect."
    * Added comments and edit context data element descriptions about the feature properties.

    Signed-off-by: Benjamin Jillich <jillich@amazon.com>

commit ddad62b994be1e89cc394b28863faa44650c8ed1
Author: Benjamin Jillich <jillich@amazon.com>
Date:   Fri Jan 21 09:37:55 2022 +0100

    Motion Matching: Replaced the MotionMatchEventData with a DiscardFrameEventData and a TagEventData (#36)

    * Replaced the MotionMatchEventData with a DiscardFrameEventData and a TagEventData
    * Moved the system components and modules into the EMFX::MotionMatching namespace
    * Converted all the animation assetinfos to use the new discard frame motion event.
    * A few other fixes and code cleaning.

    Signed-off-by: Benjamin Jillich <jillich@amazon.com>

commit f022ab4f3adbe5dc141998bc368a8ca9290698be
Author: Benjamin Jillich <jillich@amazon.com>
Date:   Thu Jan 20 07:29:01 2022 +0100

    Motion Matching: Refactoring [Part 4] Introduced MotionMatchingData (#35)

    * Renamed MotionMatchingConfig into MotionMatchingData while removing the feature schema from it as that needs to be part of the anim graph node in order to be reflected by the edit context so that users can change the features used in the algorithm.
    * Default feature schema is applied in the anim graph node in case none got de-serialized along with the node (e.g. when creating a new motion matching node).
    * Added a few class descriptions.
    * Fixed the lowest cost search frequency, which was used as the inverse (the time interval) while the UI and variable was named frequency.
    * Removed the hard-coded weighting for the past and future trajectory and created new members for the cost factors in the trajectory feature.

    Signed-off-by: Benjamin Jillich <jillich@amazon.com>

commit e05b96b0eb734cd4818e343d5b46dbb54712faf0
Author: Benjamin Jillich <jillich@amazon.com>
Date:   Wed Jan 19 10:36:20 2022 +0100

    Motion Matching: Architecture and Feature Schema Diagrams and README pass (#34)

    * Added architecture diagram
    * Added feature schema diagram
    * First pass on ReadMe.md

    Signed-off-by: Benjamin Jillich <jillich@amazon.com>

commit d50deb1f9eea79e9c56b3ded1bda62f755e3e97f
Author: Benjamin Jillich <jillich@amazon.com>
Date:   Wed Jan 19 10:35:44 2022 +0100

    Motion Matching: Refactoring [Part 3] Moved feature schema, matix and the kd-tree to the config and removed the feature database (#33)

    * Removed feature database which contained the feature schema, feature matrix and the kd-tree.
    * Feature schema, feature matrix and the kd-tree are now part of the motion matching config.
    * DebugDraw moved from the config to the instance.
    * FindLowestCostFrameIndex moved from the config to the instance.
    * Kd-tree now has a helper to calculate the number of dimensions based on a feature set.
    * SaveToCsv moved from the feature database to the feature matrix directly.

    Signed-off-by: Benjamin Jillich <jillich@amazon.com>

commit 09aff543822a2a3e092edc65cfca315a83e25730
Author: Benjamin Jillich <jillich@amazon.com>
Date:   Fri Jan 14 17:08:18 2022 +0100

    Motion Matching: Refactoring [Part 2] - Added new default feature schema and removed hard-coded locomotion config (#32)

    * Added a new FeatureSchemaDefault class that creates the default feature schema (left/right foot position and velocity, pelvis velocity and root trajectory).
    * Removed the LocomotionConfig and moved the functionality like the FindLowestCostFrameIndex() to the motion matching config.
    * Moved more per-instance data to the motion matching instance where they actually belong like the temporary cost vectors or the cached trajectory feature pointer.
    * Added cost factor to the feature base class so that users can weight the costs of the features and adjust them in the UI later on.
    * Removed the hard-coded cost factors from the motion matching node.
    * Using the new, customizable config rather than the hard-coded locomotion config in the motion matching node.

    Signed-off-by: Benjamin Jillich <jillich@amazon.com>

commit 8eba891a7b599d81a1ad7cbed841e342034fee9f
Author: Benjamin Jillich <jillich@amazon.com>
Date:   Thu Jan 13 08:56:26 2022 +0100

    Motion Matching: Refactoring [Part 1] (#31)

    * Storing the joint and relative to joint as strings in the feature so that we can later expose it to the UI.
    * Joint indices will be cached with initializing the features.
    * Some more code cleaning here and there.

    Signed-off-by: Benjamin Jillich <jillich@amazon.com>

commit 31e689015afaf61ededb50d4d3c5dddb61561603
Author: Benjamin Jillich <jillich@amazon.com>
Date:   Tue Jan 11 18:34:56 2022 +0100

    Motion Matching: Created feature schema class and separated funtionality out from the feature database (#30)

    * Created new feature schema class which holds the set of features involved in the motion matching search.  The schema represents the order of the features as well as their settings while the feature matrix stores the actual feature data.
    * Untangled feature database with the feature schema functionality and it now just holds a feature schema.
    * Code cleaning (removed Allocators.cpp, fixed alignment, renamed some variables.

    Signed-off-by: Benjamin Jillich <jillich@amazon.com>

commit c407959bd5dd1c38a283900815624a8103173d37
Author: Benjamin Jillich <jillich@amazon.com>
Date:   Tue Jan 11 13:11:51 2022 +0100

    Motion Matching: Eigen SDK is now optional and a simple NxN matrix equivalent is provided (#29)

    * Eigen SDK is now optional and users can opt-in manually.
    * Simple NxN matrix class provided for convenience that currently wraps all features needed to run motion matching (default).

    Signed-off-by: Benjamin Jillich <jillich@amazon.com>

commit dbd27a6ce68674e6fd3fcc9ca59c4d57884a133c
Author: Benjamin Jillich <jillich@amazon.com>
Date:   Mon Jan 10 08:54:01 2022 +0100

    Motion Matching: Added pose data for joint velocities and unified the broad-phase query filling (#28)

    * Added PoseDataJointVelocities class that extends the pose with relative to a given joint based velocities.
    * Unified the broad-phase search using the KD-tree by getting rid of the hard-coded fill query feature value calls. We're now just iterating over the features.
    * Adding the new pose data to the factory from within the MM gem.
    * Joint velocities are calculated before each motion matching search for the query pose.
    * Moved the debug draw helper from the locomotion config to the base class.
    * Some code cleaning.

    Signed-off-by: Benjamin Jillich <jillich@amazon.com>

commit 7ec4151fede3046379bdbdab86459cc3684c730c
Author: Benjamin Jillich <jillich@amazon.com>
Date:   Tue Jan 4 09:20:38 2022 +0100

    Motion Matching: Yet another round of timing improvements (#26)

    Motion Matching: Yet another timing improvement

    * Sampling the input query pose for the motion matching search algorithm for the new motion time as the motion instance has not been updated with the time delta when we do the search. So we pre-sample the future pose to not get little lags as the search was done on the last current pose, which is a time delta old already.
    * Fixed a bug: We never updated the previous motion instance while we were still blending between the poses. Basically we were blending with a static pose after switching to a new best matching frame. This increased smoothness obviously.
    * Re-enabled motion extraction delta blending now that we fixed the above bug which gave another smoothness increase.
    * Renaming MM behavior into config (#27)
    * Behavior -> MotionMatchingConfig
    * LocomotionBehavior -> LocomotionConfig
    * BehaviorInstance -> MotionMatchingInstance
    * Removed the MotionMatchSystem and cleaned up reflection
    * Moved to namespace EMotionFX::MotionMatching

    Signed-off-by: Benjamin Jillich <jillich@amazon.com>

commit e9c6d6fd74641bed90e57efeb5b65ad00ad44562
Author: Benjamin Jillich <jillich@amazon.com>
Date:   Fri Dec 10 10:52:06 2021 +0100

    Improved smoothness and increased trajectory facing direction influence (#25)

    * When switching to a new best matching frame we were lagging a frame behind when updating the motion instance time.
    * Increased the influence of the facing direction as we have two different position difference costs while only one facing direction cost.

    Signed-off-by: Benjamin Jillich <jillich@amazon.com>

commit f4a5d041a0d3c0bb747dd155a30a1ac993511db7
Author: Benjamin Jillich <jillich@amazon.com>
Date:   Wed Dec 8 08:54:16 2021 +0100

    Sampling poses for velocity feature directly from the motion data + some refactoring (#24)

    * New helper to calculate the velocity without a motion instance and sample the poses directly from the motion.
    * New helper to draw velocities other than the actual feature for a frame.
    * Now sharing the frame cost context between the position and the trajectory feature. The velocity feature can't use that yet as it needs custom velocity information. Will address that later.
    * After the changes, we were able to greatly simplify the extract features routine in the feature database as no motion instances are required anymore to extract the features from the source motions!
    * Some refactoring: Renamed some variables and classes for more sense.

    Signed-off-by: Benjamin Jillich <jillich@amazon.com>

commit 94e0ce6e6995cd4893eb5dada1119467dc26e360
Author: Benjamin Jillich <jillich@amazon.com>
Date:   Fri Dec 3 09:02:09 2021 +0100

    Sampling poses for trajectory feature directly from the motion data (#23)

    * Introduced a new SamplePose() helper to the Frame class which allows sampling a pose of the given frame without a motion instance. We can also apply a time offset to sample a pose before/after the frame which will be needed to sample trajectories or velocities as the Frame objects are sampled at 30 Hz currently. In case the offset reached the animation boundaries, the sample time will be clamped to the animation range.
    * The trajectory feature now samples the poses using the new helper directly from the motion data and does not need a motion instance object anymore which helps on the path towards multi-threading and also makes the code more readable and less error-prone.

    Signed-off-by: Benjamin Jillich <jillich@amazon.com>

commit 590ea3ddef575003f29e6d03b6db498a37da23a6
Author: Benjamin Jillich <jillich@amazon.com>
Date:   Thu Dec 2 09:20:48 2021 +0100

    Motion Matching: Motion extraction sometimes outputs zero movement and results in stuttering (#22)

    The problem was that after the motion matching algorithm determined a better matching frame and we started a cross-fade, at that given frame the motion extraction delta was outputting a zero vector, resulting in a little stutters at the mm update frequency. This happened because the exrtaction calculation was done before the motion instance time values were updated and also the previous time values weren't set correctly when switching the frame/animation.

    Signed-off-by: Benjamin Jillich <jillich@amazon.com>

commit 6d0b64fd4ce2e4b3ad696c6cc5f3935447a7782e
Author: Benjamin Jillich <jillich@amazon.com>
Date:   Wed Dec 1 17:05:49 2021 +0100

    Motion Matching: Facing direction support (#21)

    * Added facing direction to the trajectory feature based on a forward facing axis of the character asset.
    * The facing direction is calculated in model/animation space relative to the root joint of the character of the current frame.
    * As the trajectory is looking into the past/future of the current frame the facing direction is relative to the root joint of another frame, either in the future or the past.
    * The facing direction cost is the normalized difference (normalized dot product) between the current character facing direction and the expected one from the used frame in the database.
    * The trajectory history as well as the trajectory query got extended by a facing direction.
    * Added visualization for facing direction.

    Signed-off-by: Benjamin Jillich <jillich@amazon.com>

commit a552881139debfba2b489b8e23d57cb2556635c1
Author: Benjamin Jillich <jillich@amazon.com>
Date:   Wed Nov 24 11:02:03 2021 +0100

    Motion Matching: Moved velocity feature from normalized direction and speed to a scaled vector (#20)

    * More stable representation for joints that remain motionless.
    * Skipping velocity debug visualizations for zero velocities.

    Signed-off-by: Benjamin Jillich <jillich@amazon.com>

commit 5a27e1db3de3d4ac753208a3c4af9bc2115f5b40
Author: Benjamin Jillich <jillich@amazon.com>
Date:   Thu Nov 18 13:11:15 2021 +0100

    Motion Matching: Final tweaks for Phase 1 milestone (#19)

    * Added pelvis velocity feature to smooth out sudden direction changes of it.
    * Blending motion extraction delta as it results in smoother motion extraction after all of the fixes from the last weeks

    Signed-off-by: Benjamin Jillich <jillich@amazon.com>

commit c928dc5502cc9cbb4df3fccdae60039e68571c77
Author: Benjamin Jillich <jillich@amazon.com>
Date:   Tue Nov 16 12:45:13 2021 +0100

    Motion Matching: Move debug rendering to use DebugDisplayBus & some code cleaning (#18)

    * Debug drawing for LY Editor as well as Animation Editor in the new Atom viewport.
    * Ported from Atom aux geom to render backend independent debug display.
    * Renamed last occurances of frame data to feature.

    Signed-off-by: Benjamin Jillich <jillich@amazon.com>

commit 57904c55d15694e60713f90f42715c18756f52c4
Author: Benjamin Jillich <jillich@amazon.com>
Date:   Thu Nov 11 16:38:22 2021 +0100

    Motion Matching: Move control spline into separate trajectory query class (#17)

    * Introducing new TrajectoryQuery class which represents the trajectory history and desired future trajectory control points input for the motion matching search.
    * The trajectory history will be used to sample the number of expected control points for the past trajectory based on the trajectory feature.
    * For the future trajectory, I just ported over the different functions that we had previously. Will revisit that with a later change.
    * Moved DebugDrawControlSpline() from locomotion behavior to the TrajectoryQuery.
    * Trajectory query is owned by the behavior instance, same as the trajectory history.

    Signed-off-by: Benjamin Jillich <jillich@amazon.com>

commit fbed5b1803f820a86ffa562a2be9bbae00c64109
Author: Benjamin Jillich <jillich@amazon.com>
Date:   Tue Nov 9 10:36:03 2021 +0100

    Motion Matching: Trajectory feature improvements (#16)

    * Reduced dimensionality of the trajectory feature. Removed the velocity from the sample as the velocity is embedded in the distances between the control points already.
    * Reduced the position and facing direction features from 3D to 2D as motion extraction is projected to the ground plane anyway and the last component was always zeroed out.
    * Improved the cost function by removing the angle difference from it. After the latest changes in the trajectory history and the bug fixes in the trajectory feature and history, results are better without it.
    * Adjusted some target path generation parameters to get more variation in the locomotion results.
    * Some minor code cleanup.

    Signed-off-by: Benjamin Jillich <jillich@amazon.com>

commit 796b763d676380e0b9fbe0ae72824a7f29e59eab
Author: Benjamin Jillich <jillich@amazon.com>
Date:   Thu Nov 4 11:58:39 2021 +0100

    Major rewrite of the trajectory history (#15)

    * Fixed several sampling issues related to indexing issues by elevating the keytrack class that has been proven to work rather than reinventing the wheel here.
    * Offers two ways to sample the trajectory history, either based on time in seconds or using a normalized value.
    * Automatically takes care that the trajectory feature is not needing a longer history than the history itself records.
    * History older than the trajectory feature requres is rendered as semi-transparent spheres that fade out over time.
    * Trajectory history is now owned by the behavior instance rather than the anim graph node to have everything at a central place.
    * Trajectory history is pre-filled with the character location at init time so that we simulate a standing still character without confusing the motion matching algorithm or special case handling.

    Signed-off-by: Benjamin Jillich <jillich@amazon.com>

commit 80c6572380e1b879014aae5aecdaaea1aa3160c1
Author: Benjamin Jillich <jillich@amazon.com>
Date:   Wed Nov 3 09:34:16 2021 +0100

    Motion Matching: Improve trajectory feature extraction and cost calculation (#14)

    * Fixed bug in calculating the past frame index which led to retrieving wrong data and skewed costs.
    * Fixed an off-by-one indexing issue. We were only seeing 5 viual control points in our debug viz while there should have been 6. That is now fixed + the cost is more accurate as we're actually comparing the right control points now.
    * Changed the cost function from just calculating the spatial differences between the desired and actual trajectory positions to a combination of relative spatial differences and their angles. This reduced the average costs and resulted in better matching frames and smoother synthesized animations.

    Signed-off-by: Benjamin Jillich <jillich@amazon.com>

commit 7cc180054ede4fab8ba592c45c2feaf746beeb28
Author: Benjamin Jillich <jillich@amazon.com>
Date:   Fri Oct 29 08:54:41 2021 +0200

    Remove bad animation ranges with discard range events (#12)

    We got a mechanism in place to discard specific frames/ranges of an animation and exclude it from the motion matching database. We have several sections in our animations where the arms are over the head, where the person suddenly crouches for a bit or other poses that do not match the other locomotion data. I went through the animation database and discarded all of these sections in order to improve the results of the synthesized animations.

    Signed-off-by: Benjamin Jillich <jillich@amazon.com>

commit 14f36cdc1813aaa9a5ce514272cef01bad3f38b6
Author: Benjamin Jillich <jillich@amazon.com>
Date:   Fri Oct 29 08:54:13 2021 +0200

    Remove direction feature (facing direction will be part of the trajectory) (#13)

    Removed the currently unused and half implemented direction feature. The facing direction will be part of the trajectory in the future.

    Signed-off-by: Benjamin Jillich <jillich@amazon.com>

commit a8e6f1c42f92cb3a207bf04371e884d8f5d0fbb4
Author: Benjamin Jillich <jillich@amazon.com>
Date:   Tue Oct 26 13:30:07 2021 +0200

    Motion Matching: Improved spatial velocity feature calculation (#11)

    * Added Motion Matching ImGuiMonitor and Bus

    Added bus for pushing values to the histogram and a monitor owning and rendering the histograms.

    * Implemented performance and feature cost histograms

    * Improved spatial velocity feature calculation

    Signed-off-by: Benjamin Jillich <jillich@amazon.com>

commit 925e05bdd3c433af767118eff7b3c93bb6e2a94a
Author: Benjamin Jillich <jillich@amazon.com>
Date:   Mon Oct 25 10:26:37 2021 +0200

    Add position feature visualization and other debug viz improvements (#10)

    * Rendering a sphere for the position feature in order to see the offset between the actual foot position and the best matching frame's extracted position feature.
    * Improved velocity feature visualization by making rendering an arrow head and a thicker direction.
    * Added the ability to have different bar colors for different histograms.
    * Matched histogram bar with feature visualization colors to easily understand which 3D viz belongs to which histogram.
    * Using better matching color palette.

    Signed-off-by: Benjamin Jillich <jillich@amazon.com>

commit 6600abcff16df4e49a744bbeb68d00d9fe03f31d
Author: Benjamin Jillich <jillich@amazon.com>
Date:   Wed Oct 20 17:18:05 2021 +0200

    Use feature matrix to generalize FillFrameFloats() and improve feature visualizations (#9)

    * Generalized the FillFrameFloats() in the KD-tree by utilizing the feature matrix and dimensionality information from the features to replace the custom per-feature functions with a shared version.
    * Improved the trajectory visualization by consolidating the past and future trajectory rendering into a single function and nicer visuals by replacing the line based markers with spheres and cylinders to fake thicker line until the aux geom is able to render them.
    * Now using Atom's aux geom for improved rendering of the features.
    * Included some more feedback from the last PR.

    Signed-off-by: Benjamin Jillich <jillich@amazon.com>

commit 96a90bc62cf4220d1afd207fce9c6cab9e453c3b
Author: Benjamin Jillich <jillich@amazon.com>
Date:   Wed Oct 20 09:04:07 2021 +0200

    Motion Matching: Store features used for kd-tree in feature database and get rid of the local flags inside the feature class (#8)

    * The list of features used in the KD-tree is now separated and not part of the actual feature descriptor anymore
    * Renamed frame floats to query features / feature vales to make it align better with the rewrite from some weeks ago.
    * Some more code cleanup

    Signed-off-by: Benjamin Jillich <jillich@amazon.com>

commit e430637f734b20e6813a5c44572645390fe77c92
Author: Benjamin Jillich <jillich@amazon.com>
Date:   Fri Oct 15 09:25:49 2021 +0200

    Feature cost and performance metrics visualization (#7)

    * Added Motion Matching ImGuiMonitor and Bus

    Added bus for pushing values to the histogram and a monitor owning and rendering the histograms.

    Signed-off-by: Benjamin Jillich <jillich@amazon.com>

    * Implemented performance and feature cost histograms

    Signed-off-by: Benjamin Jillich <jillich@amazon.com>

commit f8ca765fcc168942c361dfea249b08915dac5f77
Author: Benjamin Jillich <jillich@amazon.com>
Date:   Mon Oct 11 09:21:17 2021 +0200

    Motion matching: Data analysis and Visualization (Part 2) (#6)

    Added scatterplot using PCA
    Added feature correlation heatmap
    Data normalization ground truth using sklearn's scaler (Min-max scaling)
    Histogram for verifying that the value distributions stayed the same after normalizing
    PCA scatterplot for the normalized data

    Signed-off-by: Benjamin Jillich <jillich@amazon.com>

commit 21d3f8b9e4b2545bfe07bf2a08ed1e39337dfc58
Merge: 58abd6c a6587bb
Author: Benjamin Jillich <jillich@amazon.com>
Date:   Thu Oct 7 09:16:39 2021 +0200

    Motion Matching: Feature matrix CSV export

    Signed-off-by: Benjamin Jillich <jillich@amazon.com>

commit a6587bb9218e2c618ac3a836b6374ab16fcc9bbc
Merge: 3f5ac80 aee9039
Author: Benjamin Jillich <jillich@amazon.com>
Date:   Thu Oct 7 09:13:32 2021 +0200

    Motion Matching: Start of a Jupyter notebook for feature data analysis and visualizations

    Signed-off-by: Benjamin Jillich <jillich@amazon.com>

commit aee903942ed0b05d9397778d606775482dc1cfc7
Author: Benjamin Jillich <jillich@amazon.com>
Date:   Wed Oct 6 15:35:14 2021 +0200

    Motion Matching: Feature matrix CSV export

    Signed-off-by: Benjamin Jillich <jillich@amazon.com>

commit 3f5ac808323a7b6a78cdf5c1f8ceaf8d175e9857
Author: Benjamin Jillich <jillich@amazon.com>
Date:   Tue Oct 5 10:04:44 2021 +0200

    Motion Matching: Feature matrix CSV export

    * Added GetDimensionName() function for the feature to output a component's name, which corresponds to a column in the feature matrix.
    * Added SaveAsCsv() function to feature matrix which exports a Eigen matrix to a .csv file plus column names based on the feature component names.

    Signed-off-by: Benjamin Jillich <jillich@amazon.com>

commit 58abd6c9600f67121eaec9b80c2264bba73ff198
Merge: 1fb12e4 fd12fda
Author: Benjamin Jillich <jillich@amazon.com>
Date:   Mon Oct 4 08:44:25 2021 +0200

    Motion Matching: Created feature matrix and moved all feature data to it

    Signed-off-by: Benjamin Jillich <jillich@amazon.com>

commit fd12fda7843ec6d361caa3a26892fcdb0dfc9f7c
Author: Benjamin Jillich <jillich@amazon.com>
Date:   Fri Oct 1 09:01:24 2021 +0200

    Created feature matrix and moved all feature data into it

    We now have a new FeatureMatrix which is responsible for storing the feature data that we need for the motion matching algorithm in a cache-efficient way.
    The great thing is that this is also the enabler for any feature analysis and later on machine learning as we're in the right data format already.

    The FeatureMatrix internally stores the data in a 2D dense matrix from the Eigen library. This can easily be replaced with another linear algebra/vector/matrix library though and the FeatureMatrix acts as a wrapper.

    We're currently extracting the following motion features from our motion database:
    Left foot position/velocity, right foot position/velocity and the root joint trajectory with 6 past and 6 future sample positions and directions.

    Having 41203 keyframes/poses in our motion database, this results in a feature matrix holding 22.63 MB of data. This is only the data size for the extracted features.

    Signed-off-by: Benjamin Jillich <jillich@amazon.com>

commit 7cdd4d9b0b525b3a7c218ce4a16dfa9d2a90132c
Author: Benjamin Jillich <jillich@amazon.com>
Date:   Wed Sep 22 14:21:14 2021 +0200

    Adding profile instrumentations

    Signed-off-by: Benjamin Jillich <jillich@amazon.com>

commit 1fb12e4af1fd1d70c8f39842a0207de96bd1550c
Author: Benjamin Jillich <jillich@amazon.com>
Date:   Fri Sep 17 13:10:55 2021 +0200

    Fixing compile issues due to new warning mode

    Signed-off-by: Benjamin Jillich <jillich@amazon.com>

commit 9ea3a67aa985d872498fcc19bb973311089f9679
Merge: 5e5d69c 99cfb29
Author: Benjamin Jillich <jillich@amazon.com>
Date:   Wed Sep 15 17:49:09 2021 +0200

    Added velocity feature visualization improved velocity calculation

    Signed-off-by: Benjamin Jillich <jillich@amazon.com>

commit 99cfb2914cdadcf5616570c30a15b246abb20f97
Author: Benjamin Jillich <jillich@amazon.com>
Date:   Wed Sep 8 16:35:12 2021 +0200

    Addressed PR feedback

    Signed-off-by: Benjamin Jillich <jillich@amazon.com>

commit af291c0ca97e17d1bb07075a5546747591812e9e
Author: Benjamin Jillich <jillich@amazon.com>
Date:   Tue Sep 7 10:36:18 2021 +0200

    Added velocity feature visualization improved velocity calculation

    Signed-off-by: Benjamin Jillich <jillich@amazon.com>

commit 5e5d69c9293c3609950e55d19ea6145a21027187
Merge: f2c1577 902179c
Author: Benjamin Jillich <jillich@amazon.com>
Date:   Mon Sep 6 18:58:59 2021 +0200

    MotionMatching: Separated features from source data and renamed FrameData to Feature #1

    Signed-off-by: Benjamin Jillich <jillich@amazon.com>

commit 902179c6a19f91bca64c37a383adc776da15f8e8
Author: Benjamin Jillich <jillich@amazon.com>
Date:   Fri Sep 3 15:25:27 2021 +0200

    Separated features from source data and renamed FrameData to Feature

    Signed-off-by: Benjamin Jillich <jillich@amazon.com>

commit f2c1577febb46d8957693826d404f8cc6ad5e240
Author: Benjamin Jillich <jillich@amazon.com>
Date:   Mon Aug 30 12:10:30 2021 +0200

    Fixed some compile errors for the latest version

    Signed-off-by: Benjamin Jillich <jillich@amazon.com>

commit 498018553ccf75b0e3983895984401680ffa25b9
Author: Benjamin Jillich <jillich@amazon.com>
Date:   Mon Aug 30 11:51:21 2021 +0200

    Initial motion matching prototype code

    Signed-off-by: Benjamin Jillich <jillich@amazon.com>

Signed-off-by: Benjamin Jillich <jillich@amazon.com>
 4 years ago