Are you planning to support Yolo11 anytime soon?

  • jakaskerl replied to this.

  • Hi @Thor,

    it's done, tools now support conversion of object detection YOLO11 models as well! 🙂

    Kind regards
    Jan

    Hi @Thor,

    yes, it's planned. We are working on the support as we speak. We'll let know as soon as it's ready.

    Best,
    Jan

    4 days later

    Hi @Thor,

    it's done, tools now support conversion of object detection YOLO11 models as well! 🙂

    Kind regards
    Jan

      JanCuhel I've tried the yolov11 conversion results into the yolov10 code, and the resulting fps is much slower. is there anything i missed?

        Hi @Iqbalalwi,

        I apologize, but I am not sure at what exactly you are referring. Do you mean that you compared the latency of YOLOv11 against YOLOv10 and that YOLOv10 was much faster?

        Best,
        Jan

          How much slower are we talking about? We measured various versions' latency and reported the results here. Based on our measurement, we can see that some YOLOv11 models are a bit slower than YOLOv10, but some perform slightly faster YOLOv10.

          Best,
          Jan

          2 months later

          I was experimenting with the pre-trained YOLO11n detection model from Ultralytics, and I found that if I convert it into a Blob like this:

          $ python3 -c 'from ultralytics import YOLO ; model = YOLO("yolo11n.pt") ; model.export(format="onnx")'
          $ python3 -m blobconverter --onnx-model yolo11n.onnx --shaves 3

          ...and then try to run the model using the YOLOv8 example code, I get the following error over and over again:

          [1944301041162F1200] [1.3] [3.861] [DetectionNetwork(1)] [error] Mask is not defined for output layer with width '8400'. Define at pipeline build time using: 'setAnchorMasks' for 'side8400'.

          If however I use the online model exporter to convert the weights file yolo11n.pt into a blob, then it works.

          Why is that? Is there some missing blobconverter option that the online tool applies automatically behind the scenes?

          Hi @xperroni,

          yes, that's a good question. Tools (our online exporter tailored for YOLO conversion) actually uses a slightly edited Detection head, which removes some of the final processing operations from the original head. That is because these steps are done inside the YoloDetectionNetwork node from DepthAI library, so that is why it didn't work.

          Kind regards,
          Jan