YoloX-Tracking-ncnn-RPi_64-bit icon indicating copy to clipboard operation
YoloX-Tracking-ncnn-RPi_64-bit copied to clipboard

Tflite traking fps performance

Open MyraBaba opened this issue 3 years ago • 5 comments
trafficstars

Hi, @Qengineering

I saw that below 2 link from the top is the same link and not firecting the Tflite-tracker code.

Is these fps for tflite + tracking ?

Can raspbery pi 4 handle the vehicle count ie 20 fps ?

Screen Shot 2022-08-17 at 19 45 18

MyraBaba avatar Aug 17 '22 16:08 MyraBaba

The tracking algorithm is very fast. It don't lower the inference times of the models much. That's why I have made a YoloX version with tracking, and extrapolate the other times (TFlite and NanoDet). So, TFlite with tracking will run at 20 FPS. Take a look to the YoloX code and see how the tracking is implemented. Use the same strategy for the TFlite code.

Qengineering avatar Aug 18 '22 10:08 Qengineering

@Qengineering :)

I am working monitoring the traffic on the roads and intersections etc .

  • Tflite performed poor on above and similar videos
  • what would be the best way to detect vehicles for counting purposes ?
  • And always targeted to cheap HW ie raspberryPi or ZYSJ RK3588 (https://www.cnx-software.com/2022/03/12/249-rockchip-rk3588-8k-mini-pc-16gb-ram/ very powerfull yet affordable with over priced Rp4)

Best

https://www.youtube.com/watch?v=QuUxHIVUoaY&list=PPSV

MyraBaba avatar Aug 18 '22 12:08 MyraBaba

@MyraBaba,

Please see https://github.com/Qengineering/Traffic-Counter-RPi_64-bit

Qengineering avatar Aug 31 '22 11:08 Qengineering

@Qengineering Can u please using cmake?

lucasjinreal avatar Apr 10 '23 11:04 lucasjinreal

Dear @jinfagang,

Please find the CMake and follow the instructions in the README.md

Qengineering avatar Apr 10 '23 13:04 Qengineering