Tianxiaomo / pytorch-YOLOv4

PyTorch ,ONNX and TensorRT implementation of YOLOv4
Apache License 2.0
4.47k stars 1.49k forks source link

Latency / throughput benchmarking #465

Open bkj opened 3 years ago

bkj commented 3 years ago

Hi --

Have you (or anyone here) done benchmarking on the latency and throughput of this code vs. the original.

On that page, it suggests that yolov4.cfg at 608x608 resolution should get

34(RTX 2070) FPS / 62(Tesla V100) FPS

When I run this code (either from demo.py or from models.py), I get about 33FPS on a V100.

Are there maybe performance flags that I need to set? Interested to hear other people's experience.

Thanks!