The visualization of Gaussians is strange

[yanyan-li]

Thank you for the great work. When I run the system, the Gaussian plot looks very strange. Does anyone know the cause of this problem. THX.

[muskie82]

Hi, thanks for trying our code.

Hmm, I've never seen this before.

• This is the default color rendering, right?
• Does the issue happen only in rgbd/fr3 or in all configs?

[yanyan-li]

Hi. Thanks for your reply. I tried _f3_strtexfar and _f1desk. The similar phenomenon is witnessed in these sequences. And some issues in my trying in Win10, like File "E:\MonoGS-main\utils\slam_utils.py", line 120, in get_loss_mapping_rgbd dtype=torch.float32, device=image.device RuntimeError: CUDA error: an illegal memory access was encountered CUDA kernel errors might be asynchronously reported at some other API call,so the stacktrace below might be incorrect. For debugging consider passingCUDA_LAUNCH_BLOCKING=1.`

File "E:\MonoGS-main\gaussian_splatting\utils\graphics_utils.py", line 41, in getWorld2View2 C2W = torch.linalg.inv(Rt) torch._C._LinAlgError: torch.linalg.inv: The diagonal element 1 is zero, the inversion could not be completed because the input matrix is singular. [W C:\cb\pytorch_1000000000000\work\torch\csrc\CudaIPCTypes.cpp:15] Producer process has been terminated before all shared CUDA tensors released. See Note [Sharing CUDA tensors]

Maybe some problems in my environment. I will try to find out the reason.

[muskie82]

Thanks. I think the error message looks similar to #20, can you check it?

[SClerckx]

Same issue here. Sometimes I do get splats but with unstable behaviour:

This is the console output: MonoGS: saving results in results\datasets_tum\2024-03-30-22-27-03 FEngine (64 bits) created at 000001FEB6128050 (threading is enabled) FEngine resolved backend: OpenGL MonoGS: Resetting the system MonoGS: Initialized map MonoGS: Resetting the opacity of non-visible Gaussians MonoGS: Keyframes lacks sufficient overlap to initialize the map, resetting. MonoGS: Resetting the system MonoGS: Initialized map

I have also seen the torch.linalg.inv error but it doesn't occur in every run. torch.linalg.inv: The diagonal element 1 is zero, the inversion could not be completed because the input matrix is singular.

My environment: Windows11, CUDA 11.7, Pytorch 1.12

I'm currently trying to find a lower pytorch and cuda version combo that hopefully doesn't have this problem

[foreverlong]

这里同样的问题。 有时我确实会遇到一些问题，但不稳定：

这是控制台输出： MonoGS: saving results in results\datasets_tum\2024-03-30-22-27-03 FEngine (64 bits) created at 000001FEB6128050 (threading is enabled) FEngine resolved backend: OpenGL MonoGS: Resetting the system MonoGS: Initialized map MonoGS: Resetting the opacity of non-visible Gaussians MonoGS: Keyframes lacks sufficient overlap to initialize the map, resetting. MonoGS: Resetting the system MonoGS: Initialized map

我还看到了 torch.linalg.inv 错误，但它不会在每次运行中发生。 torch.linalg.inv: The diagonal element 1 is zero, the inversion could not be completed because the input matrix is singular.

我的环境：Windows11、CUDA 11.7、Pytorch 1.12

我目前正在尝试找到一个较低的 pytorch 和 cuda 版本组合，希望不会出现这个问题

I have this issue too. Have you ever fix it?

[dumyCq]

I am using Windows 10 and I have the same issue, I am using Pytorch 1.9.0 and CUDA 11.1. If someone wants to collaborate on this, feel free to contact me.

[dumyCq]

这里同样的问题。 有时我确实会遇到一些问题，但不稳定： 这是控制台输出： MonoGS: saving results in results\datasets_tum\2024-03-30-22-27-03 FEngine (64 bits) created at 000001FEB6128050 (threading is enabled) FEngine resolved backend: OpenGL MonoGS: Resetting the system MonoGS: Initialized map MonoGS: Resetting the opacity of non-visible Gaussians MonoGS: Keyframes lacks sufficient overlap to initialize the map, resetting. MonoGS: Resetting the system MonoGS: Initialized map 我还看到了 torch.linalg.inv 错误，但它不会在每次运行中发生。 torch.linalg.inv: The diagonal element 1 is zero, the inversion could not be completed because the input matrix is singular. 我的环境：Windows11、CUDA 11.7、Pytorch 1.12 我目前正在尝试找到一个较低的 pytorch 和 cuda 版本组合，希望不会出现这个问题

I have this issue too. Have you ever fix it?

I am using Windows 10 and I have the same issue, I am using Pytorch 1.9.0 and CUDA 11.1. If someone wants to collaborate on this, feel free to contact me. I have wechat DarrenPSU