How many frames can be processed in one second

[MingRongXi]

How many frames can be processed in one second

[saswat0]

Hi @MingRongXi
Your GPU seems reasonably fine and should provide you with a decent FPS. Please ensure that you build darknet with the cuda and cudnn flags as 1.
I could get an FPS of 5 on a CPU only system (Intel i7 8th gen) and 12 with a Tesla K80 GPU (colab)

[MingRongXi]

Thank you, saswat0. I complied the darknet with the gpu and cudnn flags as 1, so the stage of vehicle-detection and ocr which realized with yolo is very fast. But the wpod-net takes a lot of time, it spends 3s or 4s processing an image. I run it with the tensorflow-gpu and keras-gpu, which are installed by the command: conda install tensorflow-gpu, conda install keras-gpu. And the speed of GPU and CPU in wpod net is almost the same.

[saswat0]

The keras implementation of wpod-net is not memory optimised and hence takes the longest time
You could try skipping frames, running your video source in a separate thread or downgrading the video quality.
Also, what's your cuda and cudnn version? I worked best with cuda 10.2 and cudnn 7.6.5 in my case

[MingRongXi]

My cuda version is 10.2, so I install the cudatoolkit and cudnn with the command: conda install cudatoolkit

[MingRongXi]

Sorry, saswat0, I clicked in a wrong place that closed the issue. My cuda version is 10.2, so I install the cudatoolkit and cudnn with the command: conda install cudatoolkit=10.2, conda install cudnn. Because the first command specifies the cudatoolkit version, so the cudnn 7.6.5 is installed. Then I install the tensorflow-gpu with conda install tensorflow-gpu. The tensorflow-gpu 1.8 is installed. But when I run the python video.py, the gpu is not used.