Strange result on mip-nerf360 dataset

[scy5335]

I run "python python_scripts/run_mipnerf360.py" on scene "garden" in mip-nerf360 dataset. And the color of gaussians looks strange and PSNR is about 6-7.

[HLinChen]

I trained the scene with the code just now. And the PSNR is about 26. This is my rendered color. Maybe there's something wrong with the preprocessing of your data.

[scy5335]

I try processed Tanks and Temples dataset and I still get bad result. Will it cause by different environments? I run the code on RTX3090 and the following is pip package addict 2.4.0 args 0.1.0 asttokens 2.4.1 attrs 24.2.0 blinker 1.8.2 certifi 2024.8.30 charset-normalizer 3.4.0 click 8.1.7 clint 0.5.1 cmake 3.30.4 comm 0.2.2 ConfigArgParse 1.7 contourpy 1.3.0 coverage 7.6.2 cycler 0.12.1 dash 2.18.1 dash-core-components 2.0.0 dash-html-components 2.0.0 dash-table 5.0.0 decorator 5.1.1 diff_gaussian_rasterization 0.0.0 docker-pycreds 0.4.0 exceptiongroup 1.2.2 executing 2.1.0 fastjsonschema 2.20.0 filelock 3.16.1 Flask 3.0.3 fonttools 4.54.1 gitdb 4.0.11 GitPython 3.1.43 GPUtil 1.4.0 idna 3.10 imageio 2.35.1 importlib_metadata 8.5.0 iopath 0.1.10 ipython 8.28.0 ipywidgets 8.1.5 itsdangerous 2.2.0 jedi 0.19.1 Jinja2 3.1.4 joblib 1.4.2 jsonschema 4.23.0 jsonschema-specifications 2024.10.1 jupyter_core 5.7.2 jupyterlab_widgets 3.0.13 kiwisolver 1.4.7 lazy_loader 0.4 lightning-utilities 0.11.7 lit 18.1.8 lpips 0.1.4 mamba 0.11.3 MarkupSafe 3.0.1 matplotlib 3.9.2 matplotlib-inline 0.1.7 mediapy 1.2.2 mpmath 1.3.0 nbformat 5.10.4 nest-asyncio 1.6.0 networkx 3.3 ninja 1.11.1.1 numpy 1.26.1 nvidia-cublas-cu11 11.10.3.66 nvidia-cuda-cupti-cu11 11.7.101 nvidia-cuda-nvrtc-cu11 11.7.99 nvidia-cuda-runtime-cu11 11.7.99 nvidia-cudnn-cu11 8.5.0.96 nvidia-cufft-cu11 10.9.0.58 nvidia-curand-cu11 10.2.10.91 nvidia-cusolver-cu11 11.4.0.1 nvidia-cusparse-cu11 11.7.4.91 nvidia-nccl-cu11 2.14.3 nvidia-nvtx-cu11 11.7.91 open3d 0.18.0 opencv-python 4.10.0.84 packaging 24.1 pandas 2.2.3 parso 0.8.4 pexpect 4.9.0 pillow 10.4.0 pip 24.2 platformdirs 4.3.6 plotly 5.24.1 plyfile 1.1 portalocker 2.10.1 prompt_toolkit 3.0.48 protobuf 5.28.2 psutil 6.0.0 ptyprocess 0.7.0 pure_eval 0.2.3 Pygments 2.18.0 pymeshlab 2023.12.post2 pyparsing 3.1.4 pyquaternion 0.9.9 python-dateutil 2.9.0.post0 pytorch3d 0.7.8 pytz 2024.2 PyYAML 6.0.2 referencing 0.35.1 requests 2.32.3 retrying 1.3.4 rpds-py 0.20.0 scikit-image 0.24.0 scikit-learn 1.5.2 scipy 1.14.1 sentry-sdk 2.16.0 setproctitle 1.3.3 setuptools 75.1.0 simple_knn 0.0.0 six 1.16.0 smmap 5.0.1 stack-data 0.6.3 sympy 1.13.3 tenacity 9.0.0 termcolor 2.5.0 threadpoolctl 3.5.0 tifffile 2024.9.20 torch 2.0.1+cu118 torchaudio 2.0.2+cu118 torchmetrics 1.4.3 torchvision 0.15.2+cu118 tqdm 4.66.5 traitlets 5.14.3 trimesh 4.4.9 triton 2.0.0 typing_extensions 4.12.2 tzdata 2024.2 urllib3 2.2.3 vcr-gaus 0.0.0.dev0 /media/common/disk3/scy/surface_reconstruction/VCR-GauS wandb 0.18.3 wcwidth 0.2.13 Werkzeug 3.0.4 wheel 0.44.0 widgetsnbextension 4.0.13 zipp 3.20.2

[HLinChen]

Can you check the initial point clouds created from colmap? Occasionally, the point cloud generated by the colmap can be a bit problematic.

[scy5335]

There may be something wrong caused by pip. I use "cd submodules/diff-gaussian-rasterization" & "pip install -e ." to reinstall diff-gaussian-rasterization module and the bug is fixed. Thanks for your reply sincerely.

[scy5335]

I try for a lot of times，the same problem occur almost every time. There is also few times that the code can generate correct results. I have no idea what causes this problem. And I don't change the initial point cloud, it's not caused by initial point clouds

[HLinChen]

Do you meet the problem in other datasets, like TNT?

[scy5335]

Yes, if the result is correct, it will work on different datasets. And it will fail on different datasets at the same time.

[scy5335]

I reinstall submodules/diff-gaussian-rasterization, but it just work for several times. If it works, it can produce good result on different datasets. Unfortunately, I can't fixed the bug just reinstalling now.

[VictorStarkSnow]

I reinstall submodules/diff-gaussian-rasterization, but it just work for several times. If it works, it can produce good result on different datasets. Unfortunately, I can't fixed the bug just reinstalling now.

Hi, I have the same problem. I cannot get the correct rendering results or export Mesh in different datasets (TNT, DTU). I configured the environment according to the README. There were no problems during the training process, but the point cloud and rendering results were wrong. How did you solve it?

[scy5335]

I reinstall submodules/diff-gaussian-rasterization, but it just work for several times. If it works, it can produce good result on different datasets. Unfortunately, I can't fixed the bug just reinstalling now.

Hi, I have the same problem. I cannot get the correct rendering results or export Mesh in different datasets (TNT, DTU). I configured the environment according to the README. There were no problems during the training process, but the point cloud and rendering results were wrong. How did you solve it?

Unfortunately, I haven't solve it so far.

[VictorStarkSnow]

Thank you for your reply, I hope the author can solve the problem

[HLinChen]

I've tested it on different machines but didn't find this bug. I will figure it out after the CVPR deadline.