gap-lab-cuhk-sz / gaustudio Goto Github PK

Hello,

thanks for your excellent work! I meet a problem when I try to compile mvs-texturing following those commands:

git clone https://github.com/nmoehrle/mvs-texturing.git
cd mvs-texturing
mkdir build && cd build && cmake ..
make (or make -j for parallel compilation)

But I meet this issue:

-- Setting build type to 'RELWITHDEBINFO' as none was specified.
CMake Error at elibs/tbb/FindTBB.cmake:187 (file):
  file failed to open for reading (No such file or directory):

    /usr/include/tbb/tbb_stddef.h
Call Stack (most recent call first):
  CMakeLists.txt:16 (FIND_PACKAGE)

System: Linux 22.04
TBB version 2021.5, it doesn't include tbb_stddef.h anymore.

Anyone meet this issue? Thanks a lot in advance!

Hi Chongjie,

Thank you so much for your insightful work. May I kindly ask if there is any plan to support Colmap-free 3d Gaussian Splatting in your studio?

Sincerely,
Chenlin

The installation package of vdbfusion under pip does not support win platform
Perhaps we need to modify its dependencies to accommodate windows

when i run texrecon ./images ./fused_mesh.ply ./results/textured_mesh --outlier_removal=gauss_clamping --data_term=area --no_intermediate_results happen that bash: texrecon: command not found

and i had run
git clone https://github.com/nmoehrle/mvs-texturing.git cd mvs-texturing mkdir build && cd build && cmake .. make (or make -j for parallel compilation)

Hi, it's an amazing work for gaussian splatting research.

It seems the input data format for mesh extraction is based on vanilla gaussian splatting with a .ply file.

I wonder if I want to use param-based output data from some 3dgs algorithms (eg. splatam), whose output is a set of 3dgs params:

    params = {
        'means3D'
        'rgb_colors'
        'unnorm_rotations'
        'logit_opacities'
        'log_scales'
    }

how to modify the scripts.

Hello, I encountered a StopIteration error when running gs-extract-mesh, but running extract_mesh works fine for extracting mesh. I would like to know if these two are the same.

由于3DGS生成的.ply点云数据周围噪点太多，导致提取的mesh质量较差，我使用CC将点云进行处理一下，例如去噪，再进行mesh提取，但是报错这个.ply文件缺少了opacity字段。有什么方法可以解决，初始点云噪点导致mesh提取质量的问题嘛？

Hi! Thanks for your great work!
I really interested in Gaussian Sky Modeling. And I wanna know when will Gaussian Sky Modeling and Sky Mask Generation Scripts be released.

Thanks for your amazing work!

i use the following command to extract the mesh:
gs-extract-mesh -m ./data/1750250955326095360_data/result -o ./output/1750250955326095360_data

But the following errors will occur

Loading trained model at iteration 10000
./output/point_cloud/iteration_10000/point_cloud.ply
Traceback (most recent call last):
  File "/home/yxiong/anaconda3/envs/gaustudio/bin/gs-extract-mesh", line 33, in <module>
    sys.exit(load_entry_point('gaustudio', 'console_scripts', 'gs-extract-mesh')())
  File "/home/yxiong/gaustudio/gaustudio/gaustudio/scripts/extract_mesh.py", line 64, in main
    pcd.load(os.path.join(args.model,"point_cloud", "iteration_" + str(loaded_iter), "point_cloud.ply"))
  File "/home/yxiong/gaustudio/gaustudio/gaustudio/models/base.py", line 68, in load
    assert len(names) == self.config["attributes"][elem]
AssertionError

Is there any solution?

Could you show an example of cameras.json file? I want to transfer a gaussian model to mesh but have no idea on how to build the cameras.json.

将GS提取成mesh，结果没有纹理信息，是黑白的mesh。我的输入就是原始GS的结果，为啥啊

Thanks for the impressive work!

For comparison purpose, I would like to ask how the SuGaR's results (Fig. 4 in your paper) for Blender dataset were obtained.

As the official code of SuGaR didn't provide such adaption (espetially dataloader and training settings), may I ask you for the implementation details? It would be greatly appreciated, if these codes can be shared.

感谢大佬的工作，太牛了。顺便问一下大佬会不会在这个repo里加入一些支持3D Gaussian 点云的编辑算法（分割，伪label，in-paint，removal等）。

Thanks for your amazing work.
i use the following command to extract the mesh:
gs-extract-mesh -m ./data/1750250955326095360_data/result -o ./output/1750250955326095360_data

but it takes too long:

Is there any solution?

Thanks for your amazing work!

 gs-extract-mesh -m ../../gaussian-splatting/output/75744664-a/ -o ./output

I used the above command and encountered this problem. How to solve it?

Loading trained model at iteration 30000
Loaded 12195953 points from ../../gaussian-splatting/output/75744664-a/point_cloud/iteration_30000/point_cloud.ply
Loading camera data from ../../gaussian-splatting/output/75744664-a/cameras.json
  0%|                                                                                                                                                                                                                                                                                                                                               | 0/300 [00:00<?, ?it/s]/root/miniconda3/envs/gaussian_splatting/lib/python3.7/site-packages/torch/functional.py:478: UserWarning: torch.meshgrid: in an upcoming release, it will be required to pass the indexing argument. (Triggered internally at  /opt/conda/conda-bld/pytorch_1659484801627/work/aten/src/ATen/native/TensorShape.cpp:2894.)
  return _VF.meshgrid(tensors, **kwargs)  # type: ignore[attr-defined]
PointCloud provided is empty
  0%|█                                                                                                                                                                                                                                                                                                                                      | 1/300 [00:01<06:04,  1.22s/it]PointCloud provided is empty
  1%|██▏                                                                                                                                                                                                                                                                                                                                    | 2/300 [00:01<02:52,  1.73it/s]PointCloud provided is empty
  1%|███▎                                                                                                                                                                                                                                                                                                                                   | 3/300 [00:01<01:57,  2.52it/s]PointCloud provided is empty
  1%|████▎                                                                                                                                                                                                                                                                                                                                  | 4/300 [00:01<01:31,  3.25it/s]

i use the following command to extract the mesh:

gs-extract-mesh -m ./data/1750250955326095360_data/result -o ./output/1750250955326095360_data

But the following errors will occur

when i run pip install torch torchvision --index-url https://download.pytorch.org/whl/cu118 later,my torch version change to 2.2.1,like this

and you requirements.txt don't have detail version

Can you help me?Thank you so much

Hi！Thank for your great project ! I wanna know,If i already have a mesh ! for example ,the mesh come from sugar.if I can get the texture with texrecon command and what i should do

The vision of this project is very exciting, and the creation of a modular Gaussian pipeline may simplify the whole process, but it seems that this will require more time for maintainers.
The previous expected release time was early May, but now May is almost past lol,close to forgot about this project

GauStudio makes use of VDBFusion (for meshing) which is current supported only under Linux.
This is the error when runninng under Windows
>pip install -r requirements.txt

ERROR: Could not find a version that satisfies the requirement vdbfusion (from versions: none) ERROR: No matching distribution found for vdbfusion

Not sure if it helps to report this issue here.

感谢大佬的精彩工作，请问一下如何在GauStudio上运行scaffold的结果呢，我按照vanilla.yml将里面的vanilla字符改为了scafflod字符后，执行如下指令后

gs-extract-mesh -m SCGS/result -o SCGS/result/output --config scaffold
代码报错

能麻烦您告知下如何解决吗

https://arxiv.org/abs/2312.00846

How should I complete the gaussian surface reconstruction after the mesh extraction by the provided command "gs-extract-mesh -m ./data/1750250955326095360_data/result -o ./output/1750250955326095360_data"? Is there any possible guidance?

Can the online visualization "gs-viewer" draw a circle around an object and then get a video of the path that rotates around the object?

貌似在安装过程中，有一个依赖项vdbfusion，它只能在linux上面安装，在Win11上面pip找不到这个包，请问有什么办法解决吗？

Hi,
can i use the rgb and depth of the nerfstudio splatfacto model pred and
use your method to extract the mesh. Is this universal?

Thanks~

你好，感谢这篇出色的工作，在生成一些小物体上有很高的质量。
但是在我生成像playroom这样的场景mesh时，会出现孔洞较多的问题，我具体做法是Original Gaussian Splatting 30000轮后，用GauS生成mesh，结果如上图所示。
再次感谢这篇出色的工作！！！

Hello,

after training the 3DGS model on my data, I tried extracting a mesh from it using your repository. However the mesh extraction process gets killed without any explanation error. Would you have an idea why this happens? It seems to always get killed at around 80%.

Here is my output:

Loading trained model at iteration 30000
Loaded 99455 points from {path/to/my/data/tpose27}/point_cloud/iteration_30000/point_cloud.ply
Loading camera data from {path/to/my/data/tpose27}/cameras.json
  0%|                                                                                           | 0/11 [00:00<?, ?it/s]/home/imc/miniconda3/envs/gaustudio/lib/python3.8/site-packages/torch/functional.py:507: UserWarning: torch.meshgrid: in an upcoming release, it will be required to pass the indexing argument. (Triggered internally at ../aten/src/ATen/native/TensorShape.cpp:3549.)
  return _VF.meshgrid(tensors, **kwargs)  # type: ignore[attr-defined]
 82%|███████████████████████████████████████████████████████████████████▉               | 9/11 [05:52<01:12, 36.49s/it]Killed```

Thanks for your great work!
How can I train 3dgs with the DTU dataset, will you support this in the future?

Thank you so much for compiling such a great repo!

As titled, the code for SparseGS has been released. We are excited to see your integration of SparseGS/FSGS. Best wishes!

Hi authors!

I encountered this issue when running gs-extract-mesh. Any solutions to this?

texrecon /images /renders/iteration_7000/fused_mesh.ply ./results/textured_mesh --outlier_removal=gauss_clamping --data_term=area --no_intermediate_results texrecon (built on Mar 21 2024, 21:02:27) Load and prepare mesh: PLY Loader: comment https://github.com/mikedh/trimesh Reading PLY: 71353 verts... 134136 faces... done. Generating texture views: No proper input scene descriptor given. A input descriptor can be: BUNDLE_FILE - a bundle file (currently onle .nvm files are supported) SCENE_FOLDER - a folder containing images and .cam files MVE_SCENE::EMBEDDING - a mve scene and embedding

ls images
100.png 114.png 128.png 141.png 155.png 169.png 182.png 196.png 209.png 222.png 236.png 24.png 263.png 277.png 290.png 303.png 317.png 330.png 344.png 358.png 371.png 385.png 399.png 411.png 51.png 65.png 79.png 92.png
101.png 115.png 129.png 142.png 156.png 16.png 183.png 197.png 20.png 223.png 237.png 250.png 264.png 278.png 291.png 304.png 318.png 331.png 345.png 359.png 372.png 386.png 39.png 412.png 52.png 66.png 7.png 93.png
102.png 116.png 12.png 143.png 157.png 170.png 184.png 198.png 210.png 224.png 238.png 251.png 265.png 279.png 292.png 305.png 319.png 332.png 346.png 35.png 373.png 387.png 3.png 413.png 53.png 67.png 80.png 94.png
103.png 117.png 130.png 144.png 158.png 171.png 185.png 199.png 211.png 225.png 239.png 252.png 266.png 27.png 293.png 306.png 31.png 333.png 347.png 360.png 374.png 388.png 400.png 414.png 54.png 68.png 81.png 95.png
104.png 118.png 131.png 145.png 159.png 172.png 186.png 19.png 212.png 226.png 23.png 253.png 267.png 280.png 294.png 307.png 32