Troubleshooting Custom YOLOv8s-Pose Model for Action Classification on Hailo8L: Issues with Inference and Custom Class Detection

General
, , ,
21

can you send your id to dm

22

you can click on my profile and there will be a chat option

23

@nina-vilela you can find the model download link in your chat

24

Hi, did you able to solve the issue. I have the custom trained Yolo classification model for 10 different classes. I manage to compiled them to hef and run on Rasp Pi. However, all the inferences showing the output from same class, which is the wrong class

26

@junsen.yee are you doing pose detection?

27

Is is multi classification model. I have figured out the issue, is due to wrong pre-processing procedures during the model conversion with DFC

28

im facing the same problem. could you please explain what mistake you did and how to rectified that @junsen.yee

29

@nina-vilela did you find any solution?

30

after i export my pt model to onnx, i write a pre-process function to make sure onnx model show the correct output with the pre-process function. Then i use that pre-process function to pre-process the calibration file rather than using the alls from hailo

31

could please explain in detail. im a beginner to this field

1 Like
33

Hi @prasaanthg,
When running the compiled model that you shared with us, we didn’t see any detections.

Could you send us the changes that you’ve made to the model script (alls file)?