edward1997104 / neural-template Goto Github PK

Thanks for the paper/code. In the paper you mentioned the discrete phase of training has only been done on 64^3 voxels. But in configs_discrete.py you shared, the starting phase is 0. Is this file the same as the one you used for the paper?

hi，great work
i have some question about the test results，the results output ply file ,when i try open it ,i see some other things
it seems like a table
is there somethong wrong,can you give me some suggestions? thanks

Hi again, I read your code es completely , now I found some problems that doesn't clear to me . first, I didn't understand the meaning of voxel and point and values.what I'm saying I know the definition but I don't know the relationship between them ! how are they collected ?

second , If I want reconstruct an image from (2D image), what routine I do? do I construct a .hd5e file? e.g. I have an image of banana , now could you tell me how can?

I'm trying to debug the pipeline for a different setting. I was wondering if you have the loss range for beginning and end of each phase (approximately) or the overall loss range

Thanks for the paper/code. But you haven't uploaded the dataset file. The one I downloaded from the link in Dataset of 'README.md' is "pretrain.zip".

Greetings!
I hope you are doing well. I really appreciate your work. I have some confusions can you.

According to the paper, T_I is generated by the function f based on Z_T. What is the outcome of this process, or what is the form of the generated data?
Is it an image, a mesh, or another neural network?
The paper describes it as a field, so how is this field defined? Given an arbitrary 3D point, how can we determine whether this point is inside or outside the surface using T_I?
2. Was the feature encoder for generating Z_T and Z_s trained in conjunction with DT-Net? Or is it a separate and independently trained network, possibly borrowed from another study?
Thanks for time and consideration.

Great work!
I wonder that how to implement the Latent Code Arithmetic, including the shape arithmetic and topological arithmetric.

Thanks for your amazing work!
When I try to run the test command, I encounter this error: "no module named models".

I don't know how to fix it.

when i run cmd: conda env create -f environment.yml
there is some error, like:
D:\Git\Neural-Template>conda env create -f environment.yml
Collecting package metadata (repodata.json): done
Solving environment: failed

ResolvePackageNotFound:

• pip==22.1.2=py38h06a4308_0
• ld_impl_linux-64==2.38=h1181459_1
• lame==3.100=h7b6447c_0
• h5py==3.6.0=py38ha0f2276_0
• readline==8.1.2=h7f8727e_1
• urllib3==1.26.9=py38h06a4308_0
• gmp==6.2.1=h295c915_3
• openssl==1.1.1q=h7f8727e_0
• pysocks==1.7.1=py38h06a4308_0
• tqdm==4.64.0=py38h06a4308_0
• libgomp==11.2.0=h1234567_1
• lz4-c==1.9.3=h295c915_1
• pillow==9.2.0=py38hace64e9_1
• freetype==2.11.0=h70c0345_0
• mkl_fft==1.3.1=py38hd3c417c_0
• x264==1!157.20191217=h7b6447c_0
• xz==5.2.5=h7f8727e_1
• libgcc-ng==11.2.0=h1234567_1
  ...

I think maybe you have forgotten to delete some additional data after second '=', like py38h06a4308_0

Hi!
Thanks for the paper and your code here.
I'm have a question about evaluation to be more precise about calculation of the metrics such as Light field distance (LFD), Chamfer distance (CD) and Point-to-surface distance (P2F) - is there any code on PyTorch which do you use to calculate metrics on your pretrained network? Or could you give some advice there I could take it?

Hi , I had been trying to run your code in Google Colab , but after I ran , without any errors kernel had stopped!
I tried to find where this problem was happened and in debugger.py in this part:

        ######## redirect the standard output
        if is_save_print_to_file:
            sys.stdout = open(self.file_path("print.log"), 'w')

            ######## print the dir again on the log
            print("=================== Program Start ====================")
            print(f"Output directory: {self._debug_dir_name}")

kernel interrupted !
Could You Help ME ?!
Thank you

For benchmarking against different datasets and comparing to https://github.com/czq142857/BSP-NET-pytorch/blob/master/main.py if I want to train a model with BSP-net only (without the flow part) I only need to change the loss function a little bit and use the flow decoder and that should work. Is there any other part that I'm missing?

Hi,Edward. Thanks for sharing the code. And I want to know the training time and the gpu you used during training.
Excellent work. thanks again.