How to Compute mAP50 After Converting YOLOv11 to FP16 TensorRT Engine in DeepStream

Accelerated Computing Intelligent Video Analytics DeepStream SDK
,
1

Setup:
• Hardware Platform (Jetson / GPU): Jetson Nano
• DeepStream Version: 6.0.1

I am running YOLOv11 successfully with DeepStream 6.0.1 on a Jetson Nano. However, after converting my YOLOv11 model to an FP16 .engine file using TensorRT, I need to compute the mAP50 (Mean Average Precision at IoU=0.5) to evaluate its performance.

Can you suggest me how to do it ?

Any guidance or references would be greatly appreciated!

3

Can you consult TensorRT team for how to measure mAP with TensorRT? DeepStream is just a deployment SDK. The TensorRT forum is Latest Deep Learning (Training & Inference)/TensorRT topics - NVIDIA Developer Forums

1 Like
4

You can just load the TensorRT model in Ultralytics and run model.val() just like normal PyTorch model.

5

Have you tried it? I tried but meeting error.

6

What’s the error?