No labels show using Custom Model (Hef) on Raspbery Pi 5

General
1

I have set up and created a custom ‘.hef’ file successfully multiple times for use with my Hailo-8l hardware on a Raspberry Pi 5. I followed all the necessary steps, converted the model and deployed it, but I am encountering an issue where no labels are appearing when running the following command. No errors are shown. Could you please assist in resolving this issue? I am using the basic_pipelines/detection.py in the hailo-rpi5-examples to test it out.

python3 detection.py --hef-path …/resources/best.hef --input rpi --labels-json …/resources/yolov5.json --show-fps

yolov5.json
{
“detection_threshold”: 0.5,
“max_boxes”:200,
“labels”: [
“background”,
“Object”
]
}

2

Hey @Sam,

Welcome aboard! We’re delighted to have you join the Hailo Community. Congratulations on successfully creating your HEFs (Hailo Executable Format files).

To help ensure everything is working correctly, I’d be happy to review your work. Please send me the following:

  1. Your HEF file
  2. The corresponding JSON file

I’ll run these on my end to verify their functionality.

HEF Verification Steps

Here are some steps you can take to check if your HEF file is properly formatted:

  1. Use the HailoRT CLI to parse the HEF:
hailortcli parse-hef {model}

Replace {model} with the path to your HEF file.
2. Visualize the HAR (Hailo Archive) model:

hailo visualizer {HAR model}

Replace {HAR model} with the path to your HAR file.

These commands will help you verify the structure and content of your files.

If you have any questions about these steps or need further assistance, please don’t hesitate to ask. We’re here to help!

4

I re-ran all the steps but got the same result.
Is there a preferred way i can send you the HEF file?

best_simplified.hef

hailortcli parse-hef best_simplified.hef 
Architecture HEF was compiled for: HAILO8L
Network group name: best_simplified, Multi Context - Number of contexts: 2
    Network name: best_simplified/best_simplified
        VStream infos:
            Input  best_simplified/input_layer1 UINT8, NHWC(640x640x3)
            Output best_simplified/conv41 UINT8, FCR(80x80x64)
            Output best_simplified/conv42 UINT8, FCR(80x80x1)
            Output best_simplified/conv52 UINT8, FCR(40x40x64)
            Output best_simplified/conv53 UINT8, FCR(40x40x1)
            Output best_simplified/conv62 UINT8, FCR(20x20x64)
            Output best_simplified/conv63 UINT8, FCR(20x20x1)

labels.json

{
    "detection_threshold": 0.5,
    "max_boxes":200,
    "labels": [
      "unlabeled",
      "cube"
    ]
}
5

Additionally how would I be able to install the hailo visualizer?

6

The Hailo Visualizer is installed within the same environment as the DFC or the Model Zoo. As for the HEF file, please send it to me via DM here that’s the best way to share it.

7

Can you solve solution for display correct labels @omria

8

@Sam plz do help to solve display correct labels

9

I have almost the same issue but mind the detect whole screen as cars

11

hailortcli parse-hef yolov8s.hef

Architecture HEF was compiled for: HAILO8L
Network group name: yolov8s, Multi Context - Number of contexts: 3
Network name: yolov8s/yolov8s
VStream infos:
Input yolov8s/input_layer1 UINT8, NHWC(640x640x3)
Output yolov8s/yolov8_nms_postprocess FLOAT32, HAILO NMS(number of classes: 9, maximum bounding boxes per class: 100, maximum frame size: 18036)
Operation:
Op YOLOV8
Name: YOLOV8-Post-Process
Score threshold: 0.200
IoU threshold: 0.70
Classes: 9
Cross classes: false
Max bboxes per class: 100
Image height: 640
Image width: 640