deepbands/buildseg

Create plugin with trained YOLOv8 segmentation model

falahatshahrzad opened this issue · 3 comments

Hello;

I trained YOLOv8 segmentation model on custom dataset. I converted it to onnx file to create a custom buildseg plugin. My onnx file can load perfectly using python codes but it can not return any result in QGIS environment. Can you guide me how can I solve this problem?

Thank you for your guidance
Have a good time

Hello Shahrzad,
Can you please provide more details about how you "create a custom buildseg plugin" and also provide the versions of python, pip packages installed, QGIS and GDAL/OGR you are using ?

I converted the trained YOLOv8 model to onnx file and upload it in buildseg. It is necessary to say that YOLOv8 is trained in Pytorch framework. I know that buildseg models are trained in Paddlepaddle framework but I don't know how I can use the trained Pytorch models in it. Also, I used Ubuntu and python3.7 for this purpose.

@falahatshahrzad I think it might have some differences in pre-processing (e.g. parameters of normalization) or post-processing, and I think you can use infer code and your model to find what's going wrong.