i want to retrain the yolov8 pose model to detect other classes such as badminton racket (with base and head keypoints) and shuttle(with base keypoint), while preserving the ability to detect person keypoints (COCO).
I have labelled images for this task using roboflow, but for the model to still detect person keypoints do I need to label that in my custom dataset as well? is there an easier way to preserve that functionality?
Do I need to make changes to any of the following:
hailomz convert command
data.yaml
postprocess so file
Please let me know if more info is needed to understand my question clearly.
Thanks
To achieve your goal of retraining the YOLOv8 pose model while preserving its ability to detect COCO person keypoints:
Dataset
Keep marking those person keypoints (like shoulders, elbows, etc.) when you label your data. This way, the model keeps its people-spotting skills while learning to find badminton rackets and shuttles too. It’s like teaching someone to walk and chew gum at the same time!
You’ll need to tweak two things:
Your data.yaml file - just update it with where your new dataset lives and add your new stuff (rackets, shuttles) alongside the person class it already knows about.
The postprocess SO file - you probably won’t need to touch this unless you’re doing something fancy with how the model spots the new objects. Just make sure it can handle all your classes and you should be good to go.