Error occurs on other data sets.
Warrior456 opened this issue · comments
Hi! Great work! The following error occurs when I run on other data sets. Do you have any suggestions for this. Thank you!
File "/data4/hule/CODE/GS/GPS-Gaussian/gaussian_renderer/init.py", line 62, in render
cov3D_precomp=None)
File "/data4/hule/opt/anaconda3/envs/gs/lib/python3.7/site-packages/torch/nn/modules/module.py", line 1194, in _call_impl
return forward_call(*input, **kwargs)
File "/data4/hule/opt/anaconda3/envs/gs/lib/python3.7/site-packages/diff_gaussian_rasterization/init.py", line 219, in forward
raster_settings,
File "/data4/hule/opt/anaconda3/envs/gs/lib/python3.7/site-packages/diff_gaussian_rasterization/init.py", line 41, in rasterize_gaussians
raster_settings,
File "/data4/hule/opt/anaconda3/envs/gs/lib/python3.7/site-packages/diff_gaussian_rasterization/init.py", line 90, in forward
raise ex
File "/data4/hule/opt/anaconda3/envs/gs/lib/python3.7/site-packages/diff_gaussian_rasterization/init.py", line 86, in forward
num_rendered, color, radii, geomBuffer, binningBuffer, imgBuffer = _C.rasterize_gaussians(*args)
RuntimeError: an illegal memory access was encountered
It seems like an error in the Gaussian renderer. First, I suggest passing the test on the provided real-world data to exclude the error caused by the environment. Then, rebuild the 3d point clouds from the estimated depth to check whether the depth map is reasonable. Afterward, check whether all parameters in GaussianRasterizationSettings and GaussianRasterizer are cuda tensors.
Thank you for your timely reply. There was no problem when I tested with the data set you provided. And all parameters in GaussianRasterizer and GaussianRasterizationSettings are cuda tensors. As for the depth, I didn't visualize it. But in theory, even if the depth estimate is not correct, I think the rendering process should not report errors. So it's confusing to me.
I just checked it again and there is indeed one variable that is not on cuda and there is no error after it is placed on cuda. Thank you so much, great job, good night!