Great job! I'm new to this field, and I want to know if it's possible to reconstruct from an RGB image?

Now that value of points_real is normalized to [0,1] in each step, why not just set the max_real to 1.0 instead of 0.003?

In the config file, for example, exp/config/exp_3/thingi32/252119.yaml. I can not find a parameter to control the final number of faces. It seems that the only relevant parameter is refine. However, the parameter refine can not directory control the face number either.

It was a nightmare trying to install it on windows, I spent about 6 hours finally find a way to make it work, following are main problems I encountered along the way:

Assume you already initialize your conda environment, and cloned this project

In windows, use cmake --build . instead of make

When build [DMeshRenderer](https://github.com/SonSang/dmesh_renderer), if you encounter error:

• fatal error C1083: Cannot open include file: 'glm/glm.hpp': No such file or directory

Solution:

1. Installing Vcpkg:

   git clone https://github.com/microsoft/vcpkg
   cd vcpkg
   .\bootstrap-vcpkg.bat
   # Add environment variable: VCPKG_ROOT = "Your vcpkg directory" e.g. C:\Users\reall\Softwares\vcpkg
   # Add same path to Path environment variable

2. Install GLM: vcpkg install glm

   Copy glm folder from vcpkg\packages\glm_x64-windows\include to your conda environment’s include folder

When build & install **[OneTBB](https://github.com/oneapi-src/oneTBB)**  if you encounter error:

• CMake Error at src/tbb/cmake_install.cmake:51 (file): file INSTALL cannot find "C:/Users/reall/Softwares/Miniconda/envs/AI3D_Exp/_Projects/dmesh/external/oneTBB/build/msvc_19.39_cxx_64_md_release/tbb12.dll": File exists.

Solution:

• oneapi-src/oneTBB#708 (comment)

  cd external/oneTBB
  mkdir build && cd build
  cmake -DCMAKE_INSTALL_PREFIX=../install -DTBB_TEST=OFF ..
  cmake --build . --config release
  cmake --install . --config release

Install [CGAL 5.6.1 - Manual: Using CGAL on Windows (with Visual C++)](https://doc.cgal.org/latest/Manual/windows.html#install-from-source)

When build cgal_wrapper, if you encounter error:

• dmesh\external\cgal\Installation\include\CGAL\config.h(111,10): error C1083: Cannot open include file: 'boost/config.hpp': No such file or directory

Solution:

• In dmesh\cgal_wrapper\CMakeLists.txt add additional line after target_include_directories() to include_directories("boost folder’s parent folder, e.g. C:/Users/reall/Softwares/boost_1_85_0")

When build DMesh (running pip install -e . in dmesh directory), If you encounter error

1. LINK : fatal error LNK1181: cannot open input file 'cgal wrapper.lib’

   Solution:

   • Move cgal wrapper.lib from dmesh\cgal_wrapper\Debug\ under dmesh\cgal_wrapper
   • Or move it to other include library path, check dmesh\setup.py

2. LINK : fatal error LNK1181: cannot open input file 'gmp.lib’ or 'mpfr.lib’

   Solution:

   • Move gmp.lib or 'mpfr.lib’ from vcpkg\packages\gmp_x64-windows\lib or vcpkg\packages\mpfr_x64-windows\lib to your conda environment’s libs folder

@SonSang

Awesome work based on a great idea. But a little suggestion, if you want people to use your work and create more useful tools or research papers with it, then you definitely want to put in some effort to make it easy to install. This project is easily one of the hardest project I ever installed.

Cheers, have a good day :)

Hello, and thank you for the excellent research. :)

I am currently testing code with custom point cloud result ( w/ only vertices ) ,
I want to do an experiment with a point cloud that does not have a ground truth (GT) mesh ( w/ vertices, faces ).

After looking at Supplementary Algorithm 2, I thought it might be possible to perform reconstruction even without the GT.
However, the code seems to be written assuming the presence of a GT mesh.

Could you please provide guidance on how to proceed in the absence of a GT mesh?
And, I'm wonder if there are any plans to release the code for experiments without GT mesh.

Thank you

Awesome!
Next, study carefully...

when run bash run_all.sh
error occurred

Traceback (most recent call last):
File "/home/code/dmesh/exp/1_mesh_to_dmesh.py", line 16, in
from diffdt import DiffDT
File "/home/code/dmesh/diffdt/init.py", line 5, in
from diffdt.pd import PDVertexComputationLayer, PDStruct
File "/home/code/dmesh/diffdt/pd.py", line 8, in
from . import _C
ImportError: /home/code/dmesh/diffdt/_C.cpython-310-x86_64-linux-gnu.so: undefined symbol: _ZN3tbb6detail2r117deallocate_memoryEPv

Thanks for your great work. I built the environment successfully following the instructions , however , I met this error when running the code :
Traceback (most recent call last):
File "/data/cz/dmesh/exp/1_mesh_to_dmesh.py", line 16, in
from diffdt import DiffDT
File "/data/cz/dmesh/diffdt/init.py", line 5, in
from diffdt.pd import PDVertexComputationLayer, PDStruct
File "/data/cz/dmesh/diffdt/pd.py", line 8, in
from . import _C
ImportError: libtbb.so.12: cannot open shared object file: No such file or directory

In section C.3.1 of the paper you mentioned that the implementation of F_A (the first renderer) does not produce visibility-related gradients (near face edges). It's hard for me to understand the term "visibility-related gradients", can you figure me out what it means in your method?

I notice that the points and weights are offloaded to cpu before delaunay triangulation, wasn't this process executed in CUDA? And why make the differential DT under torch.no_grad() context?