Slow FPS raspberry pi 5 kit

General
,
1

Only 10 FPS while trying to detect persons

Hello guys I’m newbie working with hailo8L and raspberry pi 5. Currently I’m facing a very slow rendering display while attempting to detect persons. I need to use the following coco detection model because its useful for the purpose of getting track id’s

person_detection_model_path = "yolo11n_silu_coco--640x640_quant_hailort_hailo8l_1"

person_detection_model = dg.load_model(
    model_name=person_detection_model_path,
    inference_host_address=inference_host_address,
    zoo_url=zoo_url,
    overlay_color=(0, 255, 0),
    output_class_set={"person"}
)
with degirum_tools.Display("AI Camera") as output_display:
    for inference_result in degirum_tools.predict_stream(person_detection_model, video_source):
        output_display.show(inference_result)
2

Hi @user83
Can you share the following details: inference_host_address and resolution of your input stream?

3

Hello @shashi sure.

inference_host_address = “@local”

I didn’t defined the resolution I’m just opening video source = 0 which is my usb camera

video_source = 0
with
for inference_result in degirum_tools.predict_stream(person_detection_model, video_source):

4

Hello @shashi I tried this

person_detection_model.image_width = 320
person_detection_model.image_height = 320

Without success.

5

@user83
Once a model is compiled for a certain input size, you cannot change the input size. If you want to run it with lower resolution, you need to recompile the model.

6

I’m not trying to resize it, I want to increase the performance. now is very low from 6 to 10 FPS.

Resizing was desperate attempt to increase performance

7

Hi @user83
Let us try to dig deeper to see why the performance is so low. Did you run Hailo’s benchmarking CLI to know the expected FPS? This is a good starting point for the investigation.

8

Not yet, could you tell me how to run that? sorry I’m newbie