Export YOLOV8 models from Keras to onnx

[AleksNet5]

Hello everyone!
I’ve trying to export yolov8 models to other frameworks, to onnx and to tflite. I got some point in exporting to onnx but now model has very bad ocurasy.

If I try to convert it just by tf2onnx it gives me error about YOLOV8Detection layer.

[mehtamansi29]

Hi @AleksNet5 -

Thanks for reporting the issue.

If I try to convert it just by tf2onnx it gives me error about YOLOV8Detection layer.

Can you help me with sample code for model creation or if possible share sample model and also error traceback to dig this issue more ?

[AleksNet5]

Hello!
Sorry for delay, I have successfully converted model to onnx but it works correctly only with preprocessing with TF and imageio.v3 using:
Read by imageio
Resize_with_padding with tf
Add batch dimension with tf

Otherwise if I use code from YOLOV8 outputs decoding function I got zero results

[AleksNet5]

Hi @mehtamansi29!
I think that it is core question, bc I now trying to make it work with decoding that I copied from YOLOV8 decoding functions, and with depthai. And I get zero outputs, even though I tested it in onnx and it suppose to work the same in their framework
The only thing is input parameters, I need to specify mean, scale values and channels order(rgb or bgr) for the NN input

[github-actions]

This issue is stale because it has been open for 14 days with no activity. It will be closed if no further activity occurs. Thank you.

[mehtamansi29]

Hi @AleksNet5 -

Sorry for the delay. Here you can export the Yolov8 models from other frameworks to onnx or tflite :

from ultralytics import YOLO
model = YOLO("yolov8n.pt")
#export yolo model to onnx
model.export(format="onnx", dynamic=True,  opset=13)
# export model to tflite
model.export(format="tflite")

Attached gist here fro reference.