Kai Lan

Mixed precision for CUDA

2

I am trying to use float for the preconditioner part. As far as my understanding goes, constructing the preconditioner is done on CPU, but why do we need to specify...

Performance with AMGCL

1

I hope to tune my AMGCL for better performance. For context, I am trying to solve large symmtric Poisson matrix with 3189488 unknowns: https://drive.google.com/drive/folders/1_igHqFW3HDUReTsp7Ve6bba0zuZbpbSl?usp=sharing. I used the same code as...

First run slow with CUDA

3

I am using CUDA backend. In the code, I have pretty much the same setting as the Poisson tutorial. And I tested on the same examples: https://sparse.tamu.edu/FEMLAB/poisson3Db. I set up...

Benchmarking

1

I haven't found any tutorials on benchmarking. So far I have seen setup time for AMG really low even for large matrices. I wonder whether the setup is done on...

How to generate visualization mesh real time

2

Is there a way to generate body mesh based on model output at real-time? Isn't the model output compatible with SMPL paramters and we just need a formula like `Vertices_new...

About PositionEncoding

1

In `mdm.py` file, there is a comment "not used for final model" inside class PositionEncoding. What does that mean?

A bug in quaternion.py

1

Hi, I think there is a bug in Euler to Quaternion conversion (`euler2quat` function in `quaternion.py`) for ``zyx'' ordering. I have run the following test: ``` from scipy.spatial.transform import Rotation...

Potentially a bug in quat.py

1

There seem to be a bug in quaternion to euler conversion. Here is my test: ``` e = np.array([0.1*np.pi, 0.2*np.pi, 0.5*np.pi]) q = from_euler(e, order='xyz') e1 = to_euler(q, order='xyz') q1...

I see that there is an option to finetune VAE decoder during diffusion stage. Does it offer better results than fixing the pretrained VAE? Which one was used for the...

Custom data preprocessing

3

Hi, Is it possible to preprocess custom data without Unity? It would be nice to be able to do it in a Python script.