Regarding of Yolo-World

Hi ​@Haruto Watanabe! YOLOv8m-worldv2 isn't currently supported, unfortunately. The Voyager SDK supports standard YOLOv8, YOLO11, and YOLO26 tasks (detection, segmentation, pose, etc.), but YOLO-World's CLIP-based text encoder can't be compiled for the AIPU right now. Ultralytics

VLM-style support is on the roadmap but I don’t currently have a timeline on it. And I suppose YOLO-World specifically isn't guaranteed to be included, although I will ask internally about that.

If your class list is fixed, a fine-tuned YOLOv8 might be worth further investigation? What's the use case?

Thank you for your reply.

I intend to implement a zero-shot model to handle dynamic environments where targets change frequently, enabling the system to recognize objects captured by the camera without prior specific training.

But ALL Zero-shot Learning use CLIP-based text encoder so I have to change my solutions.

Do you have any ideas of this?