• Machine Learning
  • On tools.luxonis.com, what does a model marked as (detection only) mean?

Kind of a quick, dumb question but on the yolo version drop down on tools.luxonis.com, some versions are marked as (detection only) but I don't really understand what that means. If a model is detection only, what does that restrict it from doing compared to a version without that notation?

  • Hi @WestwardWinds,

    That's actually a good question. The reason is that those models (YoloV7 and YoloV8) were, apart from object detection, also trained on different tasks, such as instance segmentation, pose estimation, and classification (in the case of YoloV8). Therefore, they offer different variants of their models. And currently, our tools only support the conversion of object detection models, and that's why we explicitly state that in the options of those models. Just a small note: we are currently working on supporting instance segmentation, oriented object detection, and classification variants of YoloV8.

    Best,
    Jan

Hi @WestwardWinds,

That's actually a good question. The reason is that those models (YoloV7 and YoloV8) were, apart from object detection, also trained on different tasks, such as instance segmentation, pose estimation, and classification (in the case of YoloV8). Therefore, they offer different variants of their models. And currently, our tools only support the conversion of object detection models, and that's why we explicitly state that in the options of those models. Just a small note: we are currently working on supporting instance segmentation, oriented object detection, and classification variants of YoloV8.

Best,
Jan

    JanCuhel That makes sense, thank you. I thought that might be the case but I wanted to double check. Before I made this post I used the conversion tool to convert a yolov8 image segmentation model to a blob without errors but I haven't had a chance to test it on my OAK D yet. If I tried to run that blob on my device will it just fail entirely or will it still provide bounding boxes and detections just no masking?

    It will work as an object detection model, providing bounding boxes and detections but no masks. I've just tested it.

    Best,
    Jan

      2 months later

      Is there any way to get a yolov8-pose model running manually?

      Or are you by any chance already working on supporting them in your converter?

        Hi @FredWerner,

        actually yes, we are working on supporting the YOLOv8 pose model. As far as I know, there's no exact ETA, but stay tuned. We will let you know!

        Regarding the manual inference, it should be possible.

        Best,

        Jan