interdigitalinc / underpressure Goto Github PK

Thanks for your great work! I got some questions about function contacts_detection_from_amass in demo.py.

1. If I got a set of 3d joints position which have different XYZ orientations from the orientations in mvnx file, should I change it to be same with the XYZ orientations in mvnx file?
2. Let's say the ground is Z = 0, if I got a set of feet positions that are around Z = - 0.5, should I change it to Z = 0?

how to integrate with PARE?

Hello! Can you provide a file to detect the contact of the amass dataset?thank you.

hello there , thanks dev for this awesome project ,i just want to know how to use the pretrained data provided , i have a bvh file for example how to clean the foot contacts in it and export it back

Thank you for your paper, which is very valuable!

I want to test the data captured by my own xsens with your training model, but I encounter several problems:

1. How is the skeleton coordinate system stored in the mvnx file defined? I only know that the z axis is upward.

2. Are the angles stored in the mvnx piece relative to the quaternion of the parent joint or the quaternion of the world? How do they arrange their xyz order?

We look forward to your reply!

i see we can get the 3D position by angle:
positions = anim.FK(angles, skeleton, trajectory, TOPOLOGY) in Cleaner()
so the angle and position are One-to-one correspondence。

while i notice Skeleton() in visualization.py：

def __update__(self, frame: int, prev_frame: int, next_frame: int, dframe: float):
	angles = util.SU2.slerp(*self.angles[[prev_frame, next_frame]], dframe)
	positions = anim.FK(angles, self._skeleton, 0, TOPOLOGY)
	if hasattr(self, "_trajectory"):
		positions += torch.lerp(*self.trajectory[[prev_frame, next_frame]], dframe)		
	if hasattr(self, "_label") and hasattr(self, "_label_traj"):
		self._label.setPos(*torch.lerp(*self._label_traj[[prev_frame, next_frame]], dframe)[:2].tolist(), 1.8)				
	for j, joint in enumerate(self._joints):
		joint.setQuat(pd.LQuaternion(*angles[j].tolist()))
		joint.setPos(*positions[j].tolist())

And after we run
-- angles = util.SU2.slerp(*self.angles[[prev_frame, next_frame]], dframe)
we should be to run it
-- torch.lerp(*self.trajectory[[prev_frame, next_frame]], dframe).
then run it to get the 3D pose
-- positions = anim.FK(angles, self._skeleton, trajectory, TOPOLOGY)

while the args is 0 not trajectory[[prev_frame, next_frame]] in torch.lerp(*self.trajectory[[prev_frame, next_frame]], dframe).
why is 0 in vis, and the result show is right?

i have a video, and get 24 smpl 3Dpoints by https://github.com/isarandi/metrabs/blob/master/docs/API.md .
video:

24 smpl 3Dpoints:
the basic is millimeters
dr8.zip

show points:

Then run with #3 (comment)

but i can not find the result show in func cleanup.
so something is wrong, for help !

Thanks for sharing the wonderful work!

I'm having some problems visualizing an fbx file in maya2022, it doesn't look like a human skeleton.

Thanks for sharing your excellent research. I wonder whether this method can infer in the 2d custom video or images. if yes, please offer some instructions. @salvetjx @damienreyt @mourotlucas