search

CarkusL / CenterPoint

Export CenterPoint PonintPillars ONNX Model For TensorRT
MIT License
204 stars 50 forks source link

Fps of TensorRT implementation #2

Closed MaxLyubimov closed 3 years ago

MaxLyubimov commented 3 years ago

Hello, @CarkusL! It's really interesting project and a lot of work has been done. I'm wondering what fps the resulting TensorRT model has. Have you evaluated the final FPS?

CarkusL commented 3 years ago

CUDA : 11.1 TensorRT : 7.2.2-1 GPU : RTX3090 Precision Mode: FP16 Max Pillar Number : 30000 Inupt Shape: (1, 10, 30000, 20) Average Inference Time(exclude preprocess and postprocess): 6.82434 ms

MaxLyubimov commented 3 years ago

Thank you for information !