No improvement in FPS- Yolov3

Hi,
I tried running Yolov3 on my system, with and without TensorRT. The GPU usage are:
Without Tensorrt - 1500Mb
With Tensorrt - 600Mb

The tensorrt manages to reduce the GPU usage (which might be due to layer fusion, if i am not wrong), but the problem is the time require to process one frame is still same. Both of them are taking 0.9sec to process one frame.

I was expecting a significant improvement in FPS, but that doesn’t happen. Kindly help me to solve this issue. Maybe i am missing something.

I am using-
Tensorrt-5.0.2
Cuda- 9.0
Python 3.5
The code is from Sample Folder

1 Like

Hello,

are you referring to https://docs.nvidia.com/deeplearning/sdk/tensorrt-developer-guide/index.html#yolov3_onnx this example?

Yes

Hello,

I tried running Yolov3 on my system, with and without TensorRT. The GPU usage are:

I am using-
Tensorrt-5.0.2
Cuda- 9.0
Python 3.5
The code is from Sample Folder

postprocessor.process elapsed time 1s process one frame

Hey @NVES,
Can you please provide any update? I am still not getting any significant improvement in FPS.

Thanks

1 Like

Anybody resolved this? I am experiencing the same thing and have found a separate topic opened for this. https://devtalk.nvidia.com/default/topic/1058192/tensorrt/slow-post-processing-up-to-a-second-per-frame-for-yolov3_onnx/post/5365906/?offset=3#5367739.

Would anyone be able to share the solution if there is one?

1 Like