• DepthAI-v2

  • IndexError: map::at, when using self converted blob file

Hi, I downloaded the most popular yolov3 on github‘https://github.com/ultralytics/yolov3’ and converted it into an openvino blob file with the tool you provided‘http://blobconverter.luxonis.com/’. The whole conversion process went without any problems. But when I replace the blob file in /depthai-python/examples/Yolo/yolo-tiny.py with my own converted blob, I get error 'detectionNetwork.setBlobPath(nnPath)
IndexError: map::at' . I can confirm that there is no problem with the blob file path setting, and how can I solve this problem? Looking forward to your reply.

  • erik replied to this.

    Hi Alex ,
    I would follow our notebook; besides training, they also contain steps for conversion/compiling and then deploying to the OAK device. We have notebooks for 4 different YOLO versions, including yolov3. I hope this helps!
    Thanks, Erik

      5 months later

      I had the same index error with a custom developed classifier based on mobilenet_v2. I solved this issue by downgrading the openvino model optimizer to the version 2021_4. More recent versions gave all the above mentioned index error.

        2 months later

        I installed, using pip, openvino 2021.4.0 and openvino-dev 2021.4.0 (pip install openvino==2021.4.0). My corresponding tensorflow version is 2.5.0. In order to obtain the openvino blob file I used the luxonis on line tool selecting the openvino version 2021.4 and model source: openvino model. The optimizing of a custom tensorflow model goes as follows:

        !mo --saved_model_dir ./folder containing the tensorflow model\

        --mean_values [127.5,127.5,127.5] \

        --scale_values [127.5,127.5,127.5]\

        --output_dir ./folder containig the bin and xml optimizer output files

        I hope this is of any help

        I'm not able to install. The terminal gives me "no matching distribution found for openvino==2021.4.0"

        My model is .pt…. There's no option for that in blobconverter tool..

        That is a PyTorch model. I have no experience with PyTorch. Try converting it to tensorflow or onnx model.

        The Blob Converter tool is giving me an error with the onnx file.

        Can you please help if you know what this means?

        [ GENERAL_ERROR ]
        /home/jenkins/agent/workspace/private-ci/ie/build-linux-ubuntu20/b/repos/openvino/inference-engine/src/vpu/graph_transformer/src/frontend/frontend.cpp:441 Failed to compile layer "ScatterND_794": unsupported layer type "ScatterNDUpdate"