Hi, I created a medium size dataset of polygons (all 4 points, 3 classes) using roboflow.
I trained a YOLOv8-obb which was back then good enough for what I wanted but for now it isn’t good enough because the polygons are not exactly rectangles in the images.
I couldn’t find a way to transform my data to a keypoint detection one so I’d be able to train a keypoint detection (which will supply to the polygon shape better).
Is it even possible? Is there a way for me to achieve my goal without labeling my whole dataset again?
Thank you in advance, your help will be highly appreciated
We don’t have a good way to do that natively in the app.
I suggest exporting the instance segmentation annotation file from Roboflow and using chatgpt to write a conversion script that converts it to our keypoint format (see all format structures here), then re-upload to Roboflow.
Hope this helps!
This topic was automatically closed 21 days after the last reply. New replies are no longer allowed.