Error when using model created from datasets using roboflow and model training with ultralytics

[Leonn714]

Hello, I am new to QGIS and Deepness. I came across this solution while researching for a solution to train AI segmentation detection. I am looking to train my own model to detect swimming pools. I have very little understanding of python and thus opted to follow the no code guide provided in the tutorial section. I am able to export from QGIS with the following settings.

The images exported perfectly. Training in Roboflow was fine and I made sure to export the dataset as TXT - YOLOv8. I traided the dataset in Ultralytics and converted the result to ONNX. Up to this point everything is running fine. But when I import the model into QGIS I get the following error.

I have tried doing it to a smaller sample set, tried exporting it to YOLOv5, but i keep getting this error. Can someone please point me into the right direction on what I am doing wrong? Very much appreciated!

L

[bartoszptak]

It looks like you are loading the YOLO model as a Segmentor.

[Leonn714]

I changed my ONNX model to detector, and it shows up fine. But does not work in the Segmentor model. Do I need to export the dataset to something else if i want to train a segmentor model? I would ideally like to have the detection to trace around an object rather than a box.

It looks like this in Detector mode. but still fails in segmentor mode. The error makes it look like it is expecting some kind of mask. If i am understanding your comment, YOLO can not be trained as segmentor model? Sorry for the newbie question.