siwensun / neural_diffeomorphic_flow--ndf Goto Github PK

Hello! Thank you so much for your repo. Its really useful.

I trained the model on the pancreas and brain surfaces (with the same config as the pancreas). I am able to reconstruct the pancreas, but the results on brain surfaces are abnormally poor(image attached).

the template learned looks like this.

the training loss is quite low of the order of 7x10^-4 (lower than the pancreas dataset with loss as 3x10^-3).

is there anything I am missing with the new dataset while reconstructing?

any help is appreciated.

Thanks
Aakash

Hello,

I really appreciate the excellent and inspiring work you presented. After reading the paper and codes, I have some questions:

1. How is the ground truth of signed distance sampled? Is it sampled from each shape, not the template? If so, how to ensure the single signed distance function can learn the representation of the template among a class of shapes?

2. I am confused about how to involve t in the ODEFunc, could you please explain a little bit more about it?

3. You clamped the predicted signed distance to [-0.1, 0.1], why should do this kind of clamp?

Thank you very much.

Hello,
I have read your paper titled “Hybrid Neural Diffeomorphic Flow for Shape Representation and Generation via Triplane,” and I found it to be an exceptional piece of work.
I am eager to delve deeper into your study, but I was unable to locate the source code. I would be extremely grateful if you could share the source code associated with your paper. Access to your code would be invaluable in aiding my understanding of your work and exploring potential extensions and applications in my research. And I assure you that it would solely be used for learning and academic purposes.
Thank you very much for considering my request. I am looking forward to your reply.

Thank you for your great work. Could you please tell when you will be able to release the preprocess data and evaluation scripts?

Hi,
thank you for the excellent work, but I ran into a problem during the reproduction process:

How is the.obj/.ply file obtained before sampling SDF points or surface points in the data preprocessing stage?
I tried to build the corresponding mesh from the mask in the original data set, but the generated results are quite different from the pre-processing results you gave. The pancreas_0001 figure below is an example, so could you please realse the preprocessing code for generating organ shapes given the center-cropped mask volumes? I want to know if I understood something wrong.

My results

Your results

Thanks a lot in advance.

Hi,

thank you for the excellent work.

I have two questions regarding the ODEFunc and NODEBlock/Warper codes.

1. The argument t in the forward function of ODEFunc is not used. But line 66 says "#Computed dynamics of point x at time t". Could you elaborate a bit on this?

2. Regarding the conditional code c in the Warper module. The Warper module has number_of_steps NODEBlock. In the first NODEBlock, ODEFunc is used. According to line 71, the conditional code is set to zero by ODEFunc. Does it mean only the first NODEBlock takes the conditional code as input, the remaining NODEBlock all take zero conditional code as input? If yes, could you explain a bit why?

Thanks a lot in advance.

Cheers

Thanks for your great work, however, I cannot open the dataset link, could you help me check again?

https://github.com/Siwensun/Neural_Diffeomorphic_Flow--NDF?tab=readme-ov-file#:~:text=The%20preprocesed%20data%20is%20here%2C%20including