dhernandez0 / stixels Goto Github PK

hi
thanks for sharing this code. I tried to run the example but got an error said "Error: invalid device function 8". I didn't know where this came from as I searched all the source code and couldn't find this line.

I am using a laptop with Nvidia 1070 gpu, system is Ubuntu 16.04.

The full output is
$ ./stixels ../example/ 128
ap_000_29-02-2016_09-00-09_000002.png
Camera Parameters -> Tilt: -0.484829 Height: 4.4689 vHor: 13 alpha_ground: 0.158384
Error: invalid device function 8
Error: invalid device function 8
Error: invalid device function 8
It took an average of 1.0762e-42 miliseconds, inf fps

thanks.

~/DeepLearning/stixels-master/build$ ./stixels ../example 128
img.png
0
Camera Parameters -> Tilt: 0.681282 Height: -11.8573 vHor: 956 alpha_ground: -0.0524078
Error: invalid device function 8
Error: invalid device function 8
Error: invalid device function 8
It took an average of 1.00893e-42 miliseconds, inf fps

hi,

I implemented the code with the available examples and it works. When i tried to implement on some dataset where left-image (pgm ) and disparity (pgm ) and I got duplicated stixels (it draws stixels correctly for first half of the image and reply on the second half) . Do I need to use the proposed gpu-sgm? thx

Hello,
I compiled your project as you shown, and when I tried to execute the example with "./stixels ../example 128" I got :
ap_000_29-02-2016_09-00-09_000002.png
Can't compute road estimation
It took an average of -nan miliseconds, -nan fps
If you do know where is the problem, thank you in advance for your answer.

I have tested the stixels code on a machine with NVIDIA Titan X, Cuda 8 and OpenCV 2.4

The code compiles, but when testing it at runtime using valgrind I get memory leaks, so the code can lead to a segmentation fault crash.
So pretty much the code works properly sometimes, other times it crashes. Valgrind gave the following reasons for a crash:

==17706== Conditional jump or move depends on uninitialised value(s)
==17706== at 0x406928: Stixels::Finish() (Stixels.cu:161)
==17706== by 0x40440E: main (main.cu:227)
==17706==
==17706== Conditional jump or move depends on uninitialised value(s)
==17706== at 0x406939: Stixels::Finish() (Stixels.cu:162)
==17706== by 0x40440E: main (main.cu:227)
==17706==
It took an average of 46.4508 miliseconds, 21.5281 fps
==17706==
==17706== HEAP SUMMARY:
==17706== in use at exit: 5,417,827 bytes in 9,480 blocks
==17706== total heap usage: 14,227 allocs, 4,747 frees, 18,739,217 bytes allocated
==17706==
==17706== LEAK SUMMARY:
==17706== definitely lost: 32,816 bytes in 1 blocks
==17706== indirectly lost: 0 bytes in 0 blocks
==17706== possibly lost: 928,302 bytes in 958 blocks
==17706== still reachable: 4,456,709 bytes in 8,521 blocks
==17706== suppressed: 0 bytes in 0 blocks
==17706== Rerun with --leak-check=full to see details of leaked memory
==17706==
==17706== For counts of detected and suppressed errors, rerun with: -v
==17706== Use --track-origins=yes to see where uninitialised values come from
==17706== ERROR SUMMARY: 2 errors from 2 contexts (suppressed: 0 from 0)

This error is a memory leak due to

delete[] m_prior_objfromground;
delete[] m_prior_objfromobj;

at line 161 and 162 of stixels.cu

It seems those pointers are initialized but never used and calling delete on them can cause a memory leak. I have commented out those two lines and the problem seems fixed.
Now the compiled code seems to run reliably, without crashing because of segmentation fault.

Hi!
I have been trying to modify the code to use it on images fed real time via a ZED camera by combining the camera API and modifying the main.cu file to read from it rather than the folders specified. However, when I run it, the code is still using the images from the folder and unable to read from the camera. Could you please point out the places where I should modify the code to achieve this? I can then go through them one by one to do the same.