How can I determine the specific content of the Hef model I have compiled?

General
1

I used Dataflow to compile my self trained YOLOv5m model into a HEF model.But when I tried to use the Hef model for inference, I encountered a problem.
Below is my YAML file.
path: C:/Users/Lenovo/Desktop/yolo_hailo/rasbot_dataset
train: train/images
val: valid/images
test: test/images
nc: 3
names:
0: robotcar
1: wheel
2: whiteboard

I am not sure what each layer of my Hef model is, I have used analysis tools to examine its specific construction, but I still do not know how to write post-processing code.

I did not compile using Model zoo, and I am not quite sure where DeGirum is useful. Although Mr. Shashi has been recommending DeGirum for various topics, my model is not in Model zoo, and I do not know how to use my own custom model for inference.

I have read a lot of obejection code on GitHub. I still don’t know how to analyze this Hef model. I am not sure what his various parameters are, nor do I know what tools to use for inference.

2

Hey @user178,

So first thing I’d do is check if your model is actually running properly on the Hailo chip. You can verify this by running:

hailortcli run <model>

This will tell you if the compilation went smoothly and also give you the FPS numbers for your model.

Once you’ve confirmed that’s working, I’d suggest trying it out with actual detection using the examples repo: GitHub - hailo-ai/hailo-rpi5-examples

From there, you’ll probably want to write your own C++ implementation with custom post-processing. I’d recommend checking out how we handle it in our apps infrastructure repo: hailo-apps-infra/cpp at main · hailo-ai/hailo-apps-infra · GitHub

Once you’ve got your post-processing sorted, just pass it to the parent class along with your HEF file and you should be good to go!

Let me know if you run into any issues with any of these steps.