an illegal memory access

[clyran]

When I run the run_docker.sh with the dataset named office3 , I got a Exception like this

frame-000000
bundling process input! Frame[0]
processInput time cost = 0.00140446 seconds.
depthSensing time cost = 0.00080369 seconds.
total time cost = 0.00228499 seconds.
	Success! frame frame-000000 added into BundleFusion.
frame-000001
bundling process input! Frame[1]
processInput time cost = 0.00240118 seconds.
depthSensing time cost = 0.000523138 seconds.
total time cost = 0.00301199 seconds.
Mesh generation time 1.88351e-05 seconds
	Success! frame frame-000001 added into BundleFusion.
frame-000002
=============================================
== Triangle Size: 36416
=============================================
bundling process input! Frame[2]
processInput time cost = 0.00226638 seconds.
depthSensing time cost = 0.000549047 seconds.
total time cost = 0.00288746 seconds.
	Success! frame frame-000002 added into BundleFusion.
frame-000003
terminate called after throwing an instance of 'ml::MLibException'
  what():  process:46: an illegal memory access was encountered:process

I don't know the reason and hope for your reply.
THANKS!

[zhanxiaopan]

+1
But I have processed 1830 frames when I meet the same kind of error, the error message is listed below.
The environment is Ubuntu18.04 + NVIDIA RTX 3090 + CUDA11.4 + 48GB RAM.

@FangGet Could you help to figure it out?

bundling process input! Frame[1830]

== Triangle Size: 1214292

terminate called after throwing an instance of 'ml::MLibException'
what(): copyCacheFrameFrom:26: invalid argument:copyCacheFrameFrom

[FangGet]

@zhanxiaopan the real-time visualization needs to ray-trace the whole scenes to output what Pangolin need, and I did this part use GPU, that's why the memory occupancy keeps increasing. By 1) disable the gui, or 2) increase the voxel size, or 3) enable the streaming mode(which only shows the sphere range of scenes) or 4)change the visualization to cpu may help(which may not be real-time).

I have a long time not check this project, there may be more efficient ways to do the visualization work. It will be checked soon.

[H-tr]

@zhanxiaopan the real-time visualization needs to ray-trace the whole scenes to output what Pangolin need, and I did this part use GPU, that's why the memory occupancy keeps increasing. By 1) disable the gui, or 2) increase the voxel size, or 3) enable the streaming mode(which only shows the sphere range of scenes) or 4)change the visualization to cpu may help(which may not be real-time).

I have a long time not check this project, there may be more efficient ways to do the visualization work. It will be checked soon.

Hi, thanks for your work and reply. I'm facing the same issue and I tried to disable the pangolin display. However, it still doesn't work. My error message is

terminate called after throwing an instance of 'ml::MLibException'
  what():  getHeapFreeCount:189: an illegal memory access was encountered:getHeapFreeCount
Aborted (core dumped)

Looking forward to your reply!

[cjm-jamie]

@H-tr Have you ever tried to revise your dockerfile or make file?
I solve it by setting the -DVISUALIZATION=OFF in Dockerfile and rebuild docker. It works for me, disable the pangolin display.