Segmentation Fault on "addDecodeTask"


We are trying out TensorRT 3.0.1 on Ubuntu 16.04.3 host, with CUDA 9. Although Video Codec SDK 8.0.14 sample for decode works, and even cudacodec Videostream in OpenCV is able to work fine. But, as soon as I try to run a TensorRT 3.0.1 sample, I get:

GDB trace for pointing to main issue:
Thread 1 “sample_detectio” received signal SIGSEGV, Segmentation fault.
0x000000000040ea80 in main (argc=16, argv=0x7fffffffdac8) at main.cpp:136
136 pDeviceWorker->addDecodeTask(cudaVideoCodec_H264);
(gdb) bt
#0 0x000000000040ea80 in main (argc=16, argv=0x7fffffffdac8) at main.cpp:136

Any guidance would be appreciated.


Do you use the DeepStream v1.5 which is released earlier this week?

If not, could you share the information about the version you used?
Please remember that DeepStream does not support video files of container formats (eg: mp4) but only raw video files (eg: h264 files).


we met this issue also, and solve this issue by call “deepStreamInit();” at very beginning…

a diff to deepstream 1.0…


Yes, this requirement was added in release 1.5.

You can find this information in our Release Notes under “API changes”:

deepStreamInit() needs to be called by samples upfront before using the library


Thank you very much. It works with 1.5. Never received any email, etc from NVIDIA about the 1.5 release.