Hi,
Yes, TensorRT support was removed from Chrono::Sensor. The CHANGELOG gives the reasoning directly:
“The TensorRT feature was removed. With PyChrono now wrapping the Chrono::Sensor module, this feature was superfluous. Users can now transfer data to Python nodes through Chrono::ROS or directly use PyChrono.”
One clarification, because that phrasing can read like a one-to-one replacement, and it isn't quite. TensorRT let you run a neural network inside the sensor's filter graph, on the GPU, during the simulation (in-loop inference). The suggested alternatives, Chrono::ROS and PyChrono, are about getting the sensor data out of Chrono so you can process it yourself. So the practical migration is:
For the common use case, grab frames and run a model in your own Python code, this is actually more flexible and is probably what most people want. What you give up is the tightly-coupled, in-the-loop GPU inference that never left the simulator. If your workflow depended on that (say, high-rate closed-loop perception running inside the sim), you'll now be doing that inference out-of-loop in Python rather than inside Chrono::Sensor.
However, if you think about it, in reality, this is how it happens – the sensor stream is processed elsewhere (before, you had access to privileged information).
Source: CHANGELOG.md, "Chrono::Sensor features and updates" section.
Dan
------------------------------------------------
Bernard A. and Frances M. Weideman Professor
NVIDIA CUDA Fellow
Department of Mechanical & Aerospace Engineering
Department of Computer Science
University of Wisconsin - Madison
4150ME, 1513 University Avenue
Madison, WI 53706-1572
------------------------------------------------
--
You received this message because you are subscribed to the Google Groups "ProjectChrono" group.
To unsubscribe from this group and stop receiving emails from it, send an email to
projectchron...@googlegroups.com.
To view this discussion visit
https://groups.google.com/d/msgid/projectchrono/fcc89acb-fba8-4edb-af45-a37e13528351n%40googlegroups.com.