Deepstream_pose_estimation doesn't show pose estimation skeleton

Accelerated Computing Intelligent Video Analytics DeepStream SDK
#1

I am testing deepstream_pose_estimation.
With provided onnx file, I see some detection.

But when I use the following two models, detection just show small white circles only, no green lines.

resnet18_baseline_att_224x224_A
densenet121_baseline_att_256x256_B

What should I change?
I managed to work with mp4 file. The original program uses h264 raw stream.
Can detection accuracy drop using mp4 format?

#3

Hey could you share your setup with us?

#4

Which set up file? Is it config file?

#5

I sent by private message all files (cpp file that can run mp4 file, models, config file and test video).
Only pose_estimation.onnx works (still not very good). The other models don’t work.

#6

Do you receive my files? Thanks

#7

I have sent my files in message. Have u received?

#8

the device/DS version.

#9

I am using Xavier. Jetpack is 4.4 and TensorRT 7 and Deepstream 5.0

Thanks

#10

I think 1st, you need to make sure if the model itself can give a better output.
2nd, if the model’s output is well, then you need to check if the input w/o DS pipeline is the same, you can dump the model’s input via DeepStream SDK FAQ - #9 by mchi

#12