CUDA error: an illegal memory access was encountered

[tingzhongyue]

Hi,
Thanks a lot for sharing your great work.
I've built the environment properly and while running the DTU dataset, I try to use the rasterization to render the image. But I got following error:
Traceback (most recent call last):
File "train.py", line 225, in
main()
File "train.py", line 150, in main
train_epoch(config,
File "/root/autodl-tmp/idea/scripts/trainer.py", line 94, in train_epoch
image = render(view, pc, background, config, device)["render"].cuda()
File "/root/autodl-tmp/idea/gaussian_renderer/init.py", line 80, in render
"visibility_filter" : radii > 0,
RuntimeError: CUDA error: an illegal memory access was encountered
Compile with TORCH_USE_CUDA_DSA to enable device-side assertions.
Any idea how to resolve?

My system & settings:
RTX3090
Linux 20.04

I hope I can get some solution for this problem!

Thank you.

[gaetan-landreau]

Hi,

Check out this issue and the corresponding PR that is linked here . Should help :)

[MouseChannel]

same
RTX3090
Linux 20.04

same error
has any solution yet?
try but not work

[MouseChannel]

Hi, Thanks a lot for sharing your great work. I've built the environment properly and while running the DTU dataset, I try to use the rasterization to render the image. But I got following error: Traceback (most recent call last): File "train.py", line 225, in main() File "train.py", line 150, in main train_epoch(config, File "/root/autodl-tmp/idea/scripts/trainer.py", line 94, in train_epoch image = render(view, pc, background, config, device)["render"].cuda() File "/root/autodl-tmp/idea/gaussian_renderer/init.py", line 80, in render "visibility_filter" : radii > 0, RuntimeError: CUDA error: an illegal memory access was encountered Compile with TORCH_USE_CUDA_DSA to enable device-side assertions. Any idea how to resolve?

My system & settings: RTX3090 Linux 20.04

I hope I can get some solution for this problem!

Thank you.

have you fix it up?

[tingzhongyue]

您好，非常感谢您分享您的出色工作。我已经正确构建了环境，并且在运行 DTU 数据集时，我尝试使用光栅化来渲染图像。但我收到以下错误： Traceback (most最近一次调用最后): File "train.py", line 225, in main() File "train.py", line 150, in main train_epoch(config, File "/root/autodl -tmp/idea/scripts/trainer.py"，第 94 行，在 train_epoch 中 image = render(view, pc, background, config, device)["render"].cuda() 文件 "/root/autodl-tmp/idea /gaussian_renderer/ init .py”，第 80 行，渲染“visibility_filter”：半径 > 0，运行时错误：CUDA 错误：遇到非法内存访问编译以TORCH_USE_CUDA_DSA启用设备端断言。知道如何解决吗？
我的系统和设置：RTX3090 Linux 20.04
我希望我能为这个问题找到一些解决方案！
谢谢。

你修好了吗？

Are you using the source code? The reason I had problems was that I got the degree of the spherical harmonics wrong.

[JrMeng0312]

same RTX3090 Linux 20.04

same error has any solution yet? try but not work

I meet the same problem. Have you fixed it up?

[MilesCool]

Perhaps just ensure that all the parameters passed to gaussian_renderer is on the same device, i also got this error when i tried to modify something, finally i found some parameters is on cpu, and i changed it to cuda, everything worked.

[tlxhlll]

same RTX3090 Linux 20.04
same error has any solution yet? try but not work

I meet the same problem. Have you fixed it up?

I have solved this problem. you just need to install the version of diff-gaussian-rasterization

[tlxhlll]

I have solved this problem. you just need to install the version of diff-gaussian-rasterization

[this233]

Perhaps just ensure that all the parameters passed to gaussian_renderer is on the same device, i also got this error when i tried to modify something, finally i found some parameters is on cpu, and i changed it to cuda, everything worked.

thank you ! it work!

[cat-fish-yu]

您好，非常感谢您分享您的出色工作。我已经正确构建了环境，并且在运行 DTU 数据集时，我尝试使用光栅化来渲染图像。但我收到以下错误： Traceback (most最近一次调用最后): File "train.py", line 225, in main() File "train.py", line 150, in main train_epoch(config, File "/root/autodl -tmp/idea/scripts/trainer.py"，第 94 行，在 train_epoch 中 image = render(view, pc, background, config, device)["render"].cuda() 文件 "/root/autodl-tmp/idea /gaussian_renderer/ init .py”，第 80 行，渲染“visibility_filter”：半径 > 0，运行时错误：CUDA 错误：遇到非法内存访问编译以TORCH_USE_CUDA_DSA启用设备端断言。知道如何解决吗？
我的系统和设置：RTX3090 Linux 20.04
我希望我能为这个问题找到一些解决方案！
谢谢。

你修好了吗？

Are you using the source code? The reason I had problems was that I got the degree of the spherical harmonics wrong.你在使用源代码吗？我有问题的原因是，我把球谐函数的次数弄错了。

Hi, how did you solve it? I have the same problem, but other methods didn't work.

[cat-fish-yu]

Hi, Thanks a lot for sharing your great work. I've built the environment properly and while running the DTU dataset, I try to use the rasterization to render the image. But I got following error: Traceback (most recent call last): File "train.py", line 225, in main() File "train.py", line 150, in main train_epoch(config, File "/root/autodl-tmp/idea/scripts/trainer.py", line 94, in train_epoch image = render(view, pc, background, config, device)["render"].cuda() File "/root/autodl-tmp/idea/gaussian_renderer/init.py", line 80, in render "visibility_filter" : radii > 0, RuntimeError: CUDA error: an illegal memory access was encountered Compile with TORCH_USE_CUDA_DSA to enable device-side assertions. Any idea how to resolve?嗨，非常感谢分享您的伟大工作.我已经正确地构建了环境，在运行DTU数据集时，我尝试使用光栅化来渲染图像。但我得到了以下错误：Traceback（most recent call last）：File“train.py“，line 225，in main（）File“train.py”，line 150，in main train_epoch（config，File“/root/autodl-tmp/idea/scripts/trainer.py“，line 94，in train_epoch image = render（view，pc，background，config，device）[“render”].cuda（）File“/root/autodl-tmp/idea/gaussian_renderer/init.py“，line 80，in render“visibility_filter”：radius> 0，RuntimeError：CUDA error：an illegal memory access was encountered Compile with TORCH_USE_CUDA_DSA to enable device-side assertions.有办法解决吗？
My system & settings: RTX3090 Linux 20.04我的系统和设置：RTX 3090 Linux 20.04
I hope I can get some solution for this problem!我希望我能找到解决这个问题的方法！
Thank you. 谢谢

have you fix it up?你修好了吗？

Have you fix it up?

[Zhiyuan-Yuan]

Hi, Thanks a lot for sharing your great work. I've built the environment properly and while running the DTU dataset, I try to use the rasterization to render the image. But I got following error: Traceback (most recent call last): File "train.py", line 225, in main() File "train.py", line 150, in main train_epoch(config, File "/root/autodl-tmp/idea/scripts/trainer.py", line 94, in train_epoch image = render(view, pc, background, config, device)["render"].cuda() File "/root/autodl-tmp/idea/gaussian_renderer/init.py", line 80, in render "visibility_filter" : radii > 0, RuntimeError: CUDA error: an illegal memory access was encountered Compile with TORCH_USE_CUDA_DSA to enable device-side assertions. Any idea how to resolve?嗨，非常感谢分享您的伟大工作.我已经正确地构建了环境，在运行DTU数据集时，我尝试使用光栅化来渲染图像。但我得到了以下错误：Traceback（most recent call last）：File“train.py“，line 225，in main（）File“train.py”，line 150，in main train_epoch（config，File“/root/autodl-tmp/idea/scripts/trainer.py“，line 94，in train_epoch image = render（view，pc，background，config，device）[“render”].cuda（）File“/root/autodl-tmp/idea/gaussian_renderer/init.py“，line 80，in render“visibility_filter”：radius> 0，RuntimeError：CUDA error：an illegal memory access was encountered Compile with TORCH_USE_CUDA_DSA to enable device-side assertions.有办法解决吗？
My system & settings: RTX3090 Linux 20.04我的系统和设置：RTX 3090 Linux 20.04
I hope I can get some solution for this problem!我希望我能找到解决这个问题的方法！
Thank you. 谢谢

have you fix it up?你修好了吗？

Have you fix it up?

have you solve it?

[2927803072]

也许只是确保传递给 gaussian_renderer 的所有参数都在同一设备上，当我尝试修改某些内容时也遇到了这个错误，最后我发现一些参数在 cpu 上，我把它改成了 cuda，一切正常。

Hello, can you talk about the specific part that was changed?