jmccormac / pySceneNetRGBD

Scripts showing how to work with the SceneNetRGBD dataset
GNU General Public License v3.0
178 stars 46 forks source link

Segmentation fault (core dumped) with Irrlicht Visualization #23

Closed arjunds closed 6 years ago

arjunds commented 6 years ago

I keep getting this error when uncommenting the Irrlicht visualization code in scenenet_room_generator.cpp. I want to be able to visualize the scene without rendering the images in renderer so this seemed like how I would do it. After a long time debugging, I still have yet to find a solution that will help me work around this error Any help would be appreciated!

jmccormac commented 6 years ago

Unfortunately I have been unable to reproduce this, so it's a bit tricky to track down. Just to be clear we are doing the same thing, to show the visualiser in scenenet_room_generator.cpp I uncommented lines 603-615 and 732-754, and commented out lines 726-728. When I did this it visualised the physics scene without a problem. I previously used irrlicht 1.8.2 without problem but on my current system I am using a slightly updated 1.8.3 version of irrlicht - can you try that version to see if it helps?

Also, have you tried compiling with the -g flag (to the CMakeLists.txt e.g. COMPILE_FLAGS "${CHRONO_CXX_FLAGS} -g") so that you can pinpoint the error with gdb (here's an example tutorial showing how to pinpoint segfaults with gdb).

arjunds commented 6 years ago

Alright so using irrlicht 1.8.3 seems to have fixed the problem, but now I just see the layout but none of the objects in it. Is this what the visualizer is supposed to do? Update: I used a couple of couts in the second while loop for visualization and it just keeps looping through the loop and doesn't actually draw anything. Below is a picture of what the visualizer shows me.

visualizer

Also how would one go about adding their own objects to the program. I looked through the textfiles/ directory and looked through all the files. From what I understand, I just need to edit train_split_filtered_model_info_and_texture.txt and wnid_sample_probabilities.txt. However, since you did this for tens of thousands of objects, do you have a script to automate it already? Thanks!