Hello, we’re doing liquid detection using image processing for a client, and we purchased a Pi5 + HQ camera along with your 26T kit.
Everything was going well during development. When we got to your environment, we started experiencing difficulties. First, we converted our .pt file to OnnX, but we couldn’t convert it to a HEF file.
We tried a Docker image on MacOS, but it didn’t work, I think it was due to the processors. We tried it on our servers, but it didn’t work. Finally, we tried it on a computer with a graphics card, and it still didn’t work. At the final stage, it told me the model we selected wasn’t compatible. I don’t understand why this is happening. Can you help me figure out what I’m doing wrong?
I see you’ve grabbed the DFC(MZ) version 5.0.0, but that one’s not compatible with Hailo8. You’ll want to go with version 2.14, 2.15, or 2.16 instead - just pick whichever matches your hailort version.
Also, just a heads up - you’ll get the best results running the compilation on an x86 PC with Linux.
Give those versions a try and let me know if you run into any issues!
I solved this problem by downloading your docker. My problem now is that although I created the hef file, it cannot predict anything. I made 2 example videos. I can follow each step. I am creating the .hef file, but when I try it, my hef file cannot predict anything. I used yolov8 and yolov11. I converted it to onnx, then converted it to hef, unfortunately, it does not capture anything. I gave 160 photos for optimization, but I still failed. I want it to at least make a prediction, but it does not. What could I be doing wrong? I can even give you the .hef and data set I created.
That should give you the FPS numbers you’re looking for.
Since you’re working with a YOLO model and if it’s using HailoRTPP, I’d suggest trying to run it through the detection app in the hailo-rpi5-examples to see how it performs there.
If that’s not how you’re planning to deploy it though, could you let me know what your deployment setup looks like? That way I can give you more targeted help.
All the errors were solved. It was incredibly easy for me to run the .hef model with the DeGirum Hailo examples by generating a .hef file in DeGirum AI Hub and it was much more successful than the .hef file I worked on for 40 minutes. They gave me my .hef file in just 2 minutes. While I was getting 15-20 fps with my own hef file, I increased it to 30 fps using PySDK. I was very pleased and thank you and the DeGirum team.
