czq142857 / nmc Goto Github PK

Hi, zhiqin

I am trying to use the noisy inputs to train the NMC model from scratch to get similar results from paper but found the loss decrease far too slow compared to clean inputs.
for bool part:
[0/400] time: 1366 loss: 0.25179240 loss_bool: 0.91642386 loss_float: 0.00000000
[1/400] time: 2282 loss: 0.25031856 loss_bool: 0.91567051 loss_float: 0.00000000
[2/400] time: 3002 loss: 0.25059867 loss_bool: 0.91625684 loss_float: 0.00000000
[3/400] time: 3594 loss: 0.25166979 loss_bool: 0.91583282 loss_float: 0.00000000
[4/400] time: 4078 loss: 0.25226465 loss_bool: 0.91645384 loss_float: 0.00000000
...
[275/400] time: 45009 loss: 0.24075049 loss_bool: 0.91662174 loss_float: 0.00000000
[276/400] time: 45091 loss: 0.24094345 loss_bool: 0.91680372 loss_float: 0.00000000
[277/400] time: 45175 loss: 0.23898503 loss_bool: 0.91793585 loss_float: 0.00000000
[278/400] time: 45257 loss: 0.23988940 loss_bool: 0.91663897 loss_float: 0.00000000
[279/400] time: 45340 loss: 0.24025825 loss_bool: 0.91807079 loss_float: 0.00000000

Does it also occur to you when you train it? How many epochs did you train for both the bool and float part? and would you like to share noisy training weights with me to see more results from it?
Great thanks!

Thanks for your excellent work!
If I only want to use eval_cd_nc_f1_ecd_ef1.py to evaluate my mesh data, what must I prepare?

Traceback (most recent call last):
File "main.py", line 494, in
import cutils
ModuleNotFoundError: No module named 'cutils'

Can help? Where I can get that module?

Thanks for the wonderful work.
As you mentioned in the training process, the shapenet is not an option since the mesh are not in closed triangles. However, it there any way that I could convert the mesh in shapenet to closed triangle mesh?
Thank you so much for your time!

What's the tool for visualization？
Thanks.

Hello,ZHIQIN CHEN
Thanks for your excellent work!
In the code, I noticed there is a comment for the batchsize "#batch_size must be 1", and feel confuse about this requirement. Is there anything special for this?

Greetings,

Previously I've discovered your work through the Neural Dual Contouring code and publication, which has been very impressive. Unfortunately, the non-manifold meshes output from it have been extremely problematic however.

I've been trying out the capabilities of Neural Marching Cubes, which produces manifold meshes. One issue (difference) I'm seeing is that it seems to perform more poorly than expected on noisy data.

I've been working with the 3_NMC_voxel pre-trained weights, and technically have been converting point cloud data into binary voxel occupancies.

Looking at your publication, there is a mention of graceful NMC performance on noisy data, when the model has been trained on noisy data. I'm wondering if in fact you haven't yet published those weights, and if indeed not, if you could share those here?

Thanks.

Hello,ZHIQIN CHEN.
Thank you for such a wonderful job!
I have recently been working on a surface reconstruction,but the use of mc does not work well.

Q : Can I use your network and weights directly to do surface reconstruction on my data?
The only data I have available are the Signed Distance Field.

Thanks!