After a lot of attempts I’m now back on JP 4.4 and my 3 cam sample works. The decrease of fps had nothing to do with the colour space patch. I must have had another configuration somehow.
I’m now using the Primary_Nano resnet10
model-file=../../../../samples/models/Primary_Detector_Nano/resnet10.caffemodel
proto-file=../../../../samples/models/Primary_Detector_Nano/resnet10.prototxt
labelfile-path=../../../../samples/models/Primary_Detector_Nano/labels.txt
having removed all FP8 stuff and configured more memory:
[property]
workspace-size=600
All three cams doing 25 fps inference with MJPEG capture input at 640x480@30fps.
This is what I wanted to achieve.
What remains is the problem with the nvidia-l54-bootloader after sudo apt upgrade (or the GUI pendant)
dpkg: error processing package nvidia-lt4-bootloader (--configure):
installed nvidia-lt4-bootloader package post-installation script subprocess returned error exit status 1
EDIT: BTW, there is unexpectedly no performance difference between the jpegparse ! nvv4l2decoder mjpeg=1 and the jpegdec ! videoconvert approach. I would have bet the first would be faster.