Compiling the samples on Ubuntu 14.04 host.
The lane-detection and freespace-detection both terminate with:
ImageStreamer(CUDA->GL)::sendImpl, cannot memcpy to array. CUDA error: an illegal memory access was encountered
Driveworks exception thrown: DW_CUDA_ERROR: ImageStreamer(CUDA->GL)::receiveImpl, cannot unmap : an illegal memory access was encountered
Before they give the warning:
DNN: Loaded TensorRT model was created by an older version of TensorRT library, or wrong TensorRT model format. TensorRT version: 2102 != Model version: 2100
For both samples it seems xxxDetector_processDeviceAsync(…) is the reason for breaking.
The precompiled samples do run without any issues, also no complaining about different TensorRT version. So I suspect that might be a reason.
How can we fix this, so laneDetector and freespaceDetector can run on the host?
Thanks and best regards,
Sebastian