Now for Performance Capture we get to move on to the body. The Meta Human rigs are quite complicated and are custom rigs. For this reason, we are using a proxy rig that will be retargeted with the cortex data to drive the controls of the Meta Human rig.
We started working on the cortex data in Motion Builder. First I opened a new file with the proxy rig in it and characterised it as a biped rig.
Then I imported in the pre-cleaned cortex data.
With the cortex data visible within the viewport, I added an actor into the scene.
The actor needed some adjustments to fit the parameters of the opticals, as the dimensions of the actress aren't the same as the standard actor dimensions.
Dr Kennedy mentioned in his tutorial that when we put our arms up during a T-Pose, our shoulders lift. However, on the motion builder actor, the shoulders don't lift when we raise the arms. This leads to difficulty and some odd looking shoulder drooping when we retarget the data onto the model. For this reason, we rotate the z angle of the shoulders to 30 degrees while adjusting the upper arms to compensate for the increased angle.
With the actor lined up as accurately as possible to the opticals (I found Muna was much smaller with a shorter torso than the actor model), I could create a marker set. With a marker set made, I connected the correct opticals in the correct locations on the actor.
Then with the opticals assigned, it was time to change the source to mapping. I actually found that I had to go back twice to change the scale of my actor because the initial mapping resulted in some very funky movements.
When I was happy with the accuracy of the movement, I changed the target character to the proxy rig and baked the data to the rig using the actor as the source. Once done, I could take the data from Motion Builder into Maya.
In Maya, we want the proxy rig to drive the custom rig. We did this by using the Human IK. The target character became the custom rig and the source became the proxy rig with the baked data.
There were initially some issues with the precision of custom rig following the exact location of the joints on the proxy rig, so to combat this we increased the Human IK blends and pulls to 1.00.
The next step was to create a zeroed out pose to refer back to when necessary at frame -10. This means all of the rotation controls on the joints were set to a value of 0 at frame -10 (when the performance actually begins).
With the zeroed out pose and the proxy rig connected as close as possible to the custom rig, I finally baked the movements of the proxy to the custom rig. This was a satisfying moment to see the body moving after looking at the face without any body movements for so long. It felt like the performance was really coming together.
Of course, as is the case with almost EVERYTHING in this subject, there is still much cleanup to be done. One of the first we focused on was the pull vectors of the knees. The vectors appeared to follow the vertical movement quite well, but fell short on their x and z movements. This caused some beautiful knee breaking.
I created a new animation layers from the selected pull vectors, renamed the layer 'knees' and then manually key framed the vectors by moving them into positions that aligned the knees with the proxy skeleton joints that I could see underneath her mesh.
Before:
After:
Because she moves her body away from the original camera position, the last thing to do for the week was to parent the FaceCam to the head control using a parent constraint. This was a fairly familiar concept from previous assignments in this course.
It's going to be even better when the hands are animated with the lip balm, but I'm excited to have the body moving from this week.
Opmerkingen