PeopleNet Transformer v2.0 Compatibility with DeepStream 7.1

[adithyaskumar235]

Hello NVIDIA Team,

We currently use PeopleNet for real-time person detection and would like to evaluate PeopleNet Transformer v2.0 as a potential upgrade.

Our environment:

• OS: Ubuntu 22.04.5 LTS

• JetPack: 6.2.1

• L4T: 36.4.4

• CUDA: 12.6

• DeepStream: 7.1

Could you please clarify:

1. Is PeopleNet Transformer v2.0 compatible with our environment?

2. Does it provide better detection accuracy, particularly in crowded scenes and under partial occlusion, compared with standard PeopleNet?

3. What FPS and latency can we expect on Jetson hardware?

4. Which ONNX variant and TensorRT configuration do you recommend?

Our goal is to reduce missed detections across three cameras while maintaining real-time performance.

Thank you for your guidance.

Best regards,
Adithya

[yuweiw]

@adithyaskumar235

1. Is PeopleNet Transformer v2.0 compatible with our environment?

We do not officially support this model in DeepStream 7.1 and started supporting this in 8.0.

@adithyaskumar235

2. Does it provide better detection accuracy, particularly in crowded scenes and under partial occlusion, compared with standard PeopleNet?

For accuracy metrics, we recommend referring to the peoplenet_transformer_v2. Please refer to the follwing part: If you are looking to re-train with your own dataset, use the following the guidelines for highest accuracy.

@adithyaskumar235

3. What FPS and latency can we expect on Jetson hardware?

You can use the data in the following section of the model’s webpage as a general reference. Real-Time Inference Performance

@adithyaskumar235

4. Which ONNX variant and TensorRT configuration do you recommend?

You can also refer to the model and code links I attached above.