BodyPose3DNet — how to get pose for multiple people in one frame?

Hi,

The BodyPose3DNet model card lists a limitation that it only supports one person in the scene, but the example output image on the same model card shows skeletons for multiple people.

I tested bodypose3dnet_vdeployable_performance_onnx_v1.0 via ONNX Runtime on an image with 4 people, and only one person’s skeleton was produced.

Is there a supported way to get pose output for multiple people in a single frame. Could you point me to a working reference example or config for this specific setup?

Thanks.

Environment:

  • Model: bodypose3dnet_vdeployable_performance_onnx_v1.0
  • DeepStream: 7.1 (container nvcr.io/nvidia/deepstream:7.1-triton-multiarch)
  • GPU: NVIDIA GeForce RTX 4060 (8GB)
  • Driver: 595.84, CUDA 13.2 (container CUDA runtime 12.6)

Use a top-down pipeline: Person Detector → crop each detected person → run BodyPose3DNet on each crop (optionally batched) → get per pose per person in each crop.