engine_builder core dumped and cannot detect ros message

[liangchen97]

Hi I try to run roslaunch tx2_fcnn_node cnn_only.launch

when i use rosrun tx2_fcnn_node fcrn_engine_builder --uff=./resnet_nonbt_shortcuts_320x240.uff --uffInput=tf/Placeholder
--output=tf/Reshape --height=240 --width=320 --engine=./test_engine.trt --fp16
return *** deserializing
Runtime created
engine created
fp16: 1
malloc_consolidate(): invalid chunk size
Aborted (core dumped)
Is this a normal result?

And when I use the message whose topic is /image, it does not read

NODES
/
static_transform_publisher_camera (tf/static_transform_publisher)
static_transform_publisher_scan (tf/static_transform_publisher)
tx2_fcnn_node (tx2_fcnn_node/tx2_fcnn_node)

ROS_MASTER_URI=http://localhost:11311

process[static_transform_publisher_camera-1]: started with pid [24268]
process[static_transform_publisher_scan-2]: started with pid [24269]
process[tx2_fcnn_node-3]: started with pid [24279]
Error generated. /dvs/git/dirty/git-master_linux/multimedia/nvgstreamer/gst-nvarguscamera/gstnvarguscamerasrc.cpp, execute:557 No cameras available
[ WARN] [1646725341.716882575]: Failed to capture frame
[ WARN] [1646725342.717331764]: Failed to capture frame

[KirillMouraviev]

Core dump in first case is not critical, because engine was also created.
For the second case, the error is "No cameras available". By default, the node captures images directly from camera. If you want to use images from topic, you should set use_camera parameter to false.

[liangchen97]

@KirillMouraviev
Thank you, this worked.
There is another question I hope you can answer for me, my input is 1280x960, maybe the output is 320x240 without modifying the size in the engine currently? How can I output images with higher precision?

[KirillMouraviev]

Different resolutions required different engines. Default engine is designed to work with 320x240 input and 320x240 output. For higher input and output resolution, you should create appropriate neural network model and corresponding engine.