-
Hello, I have trained a model and I have tested it with detect.py and it works very well. I have exported the model to ONNX format using the command line: and it produced an ONNX file. I attempt to use the ONNX file for inference:
and the bounding boxes are all moved to one side of the image, all confidences are 0.5 - this is a result of the sigmoid, obviously. I have tried removing the sigmoid, but that doesn't really help at all, but the bounding boxes are less skewed. the first result of my output looks like this:
I have also attempted the solution here: [https://github.com//discussions/2170] with no success How can I interpret the results of the inference? |
Beta Was this translation helpful? Give feedback.
Replies: 1 comment 1 reply
-
I have worked it out The model input doesn't need to be normalised. The outputs don't need to be adjusted with sigmoid function. I have another issue now, but I don't know whether it is a problem with YOLO or ORT. The final code was:
|
Beta Was this translation helpful? Give feedback.
I have worked it out
The model input doesn't need to be normalised.
The outputs don't need to be adjusted with sigmoid function.
I have another issue now, but I don't know whether it is a problem with YOLO or ORT.
The final code was: