I am trying tappas on Pi 5 and Hailo8, and then I find some example can run on Pi5, so I write the most useful wiki about multistream inference on hailo8.
Here is the wiki Link:
Here is the youtube link:
This example can be used to do like NVR ect.
We have added the ability to modify the batch size based on the new code, significantly improving the inference speed. According to the tests conducted on YOLOv8m with an input size of 640x640 and a batch size of 8, the inference speed for a video input of 720p is as follows:
| Channel Count | PCIE Gen2 Performance | PCIE Gen3 Performance |
|---|---|---|
| 1 channel stream | 39.82FPS | 76.99FPS |
| 2 channels streams | 19.86FPS | 38.21FPS |
| 4 channels streams | 8.45FPS | 16.94FPS |
| 8 channels streams | 3.85FPS | 8.15FPS |
| 12 channels streams | 2.94FPS | 5.43FPS |
Thanks a lot for your link!
I got error when running command:
./install.sh --skip-hailort --target-platform rpi5
shows …/libs/postprocesses/detection/hailo_nms_decode.hpp:68:78: error:‘HAILO_FORMAT_ORDER_HAILO_NMS_BY_CLASS’ was not declared in this scope, do you mean ‘HAILO_FORMAT_ORDER_HAILO_NMS’ ?
I found the reason of this error. There are 2 versions of hailort.h, one is under /usr/include/hailo, the other is under tappas/hailort/sources/hailort/libhailort/include/hailo, and they are not same. The ./install.sh script uses /usr/include/hailo/hailort.h, which does not have HAILO_FORMAT_ORDER_HAILO_NMS_BY_CLASS. But the other version of hailort.h have this HAILO_FORMAT_ORDER_HAILO_NMS_BY_CLASS.
Please make sure tapps version is 3.9, and try again.
Maybe you installed both hailo-all and hailort independently.
Hi jiahao,
Is there any social network group that I can join for discussion?
What my question was mentioned in here:
Thanks,
Hari