low FPS using deepstream SDK 4.0 on Jetson TX2

I installed deepstream SDK 4.0 on my jetson TX2 and wanted to see the performance. I used yolov3 app (following the steps in README file) using default parameters. but the FPS was only arount 4.5, when I used yolov3_tiny it was around 55.
what FPS should I get in this case? did I miss any step?

Thank you

Also, I’m trying to save the output video using “deepstream_app_config_yoloV3.txt” file. any help on how to do that?

Hi,

This is related to the model complexity.
YOLOv3 is much complicated that the tiny version and this also reflect to the performance.
You can check this page for the FLOPS information of each model: YOLO: Real-Time Object Detection

Here is our benchmark result for YOLOv3 on TX2 in our environment:

YoloV3     : 5 fps
YoloV3_tiny: 41.6 fps

To record the output as video, please update the sink component into video mode.
https://docs.nvidia.com/metropolis/deepstream/dev-guide/index.html#page/DeepStream_Development_Guide%2Fdeepstream_app_config.3.2.html%23wwpID0ENHA

[sink0]
enable=1
#Type - 1=FakeSink 2=EglSink 3=File
type=3
#1=mp4 2=mkv
container=1
#1=h264 2=h265
codec=1
sync=0
#iframeinterval=10
bitrate=2000000
output-file=out.mp4
source-id=0

Please noticed that Deepstream SDK support multiple sink component as well.
Thanks.