Frigate using RPi5 AI kit and YUV420p stream

General
,
1

I have a Frigate installation on a Raspberry Pi 5 with AI kit (Hailo8L). Sadly my camera is streaming using format YUV420p 300x300 instead of RGB24 300x300.

Anyone able to use hailo8l as the detectors?

2

Hey @chanaka ,

Welcome to the Hailo Community!

I have Frigate running both at my home and in the Hailo offices for testing purposes. To help you get it running with the Hailo-8L, could you please provide more details about the specific error you’re encountering?

Looking forward to helping you get Frigate up and running with the Hailo-8L!

3

Hi @omria ,

Thanks a lot for the reply.

My camera feed is a RTSP stream from a camera running OpenIPC and it’s x264 encoded.

Here’s the output from ffprobe.

[STREAM]
width=300
height=300
pix_fmt=yuv420p
[/STREAM]

I assume this is because of the format error where as per Model Zoo input resolution is 300x300x3 for ssd_mobilenet_v1 .

Do you have an idea about another model or any solution which I can apply ?

Here’s what I found on frigate logs

2024-12-01 14:34:34.774854948  [2024-12-01 14:34:34] frigate.detectors.plugins.hailo8l ERROR   : [detect_raw] HailoRTException during inference: Invalid argument. See hailort.log for more information
2024-12-01 14:34:34.779813135  [HailoRT] [error] CHECK failed - Memory size of vstream ssd_mobilenet_v1/input_layer1 does not match the frame count! (Expected 270000, got 307200)
2024-12-01 14:34:34.779881654  [HailoRT] [error] CHECK_SUCCESS failed with status=HAILO_INVALID_ARGUMENT(2)

and it does not process frames for detection.