Hello, have been learning how to use the OAK-D with a ccustom yolov4-tiny cnn. I was successful in using my custom cnn had a height and width of 416. I wanted to increase the accuracy or the cnn, so I changed the height and width to 640. Now when I run on the OAK-D I see the following error:

[14442C10B1FE41D700] [297.970] [SpatialDetectionNetwork(14)] [error] Input tensor 'inputs' (0) exceeds available data range. Data size (1228800B), tensor offset (0), size (2457600B) - skipping inference

Have I exceeded a memory limit on the OAK-D?

Thanks,
Phillip

  • erik replied to this.

    Hello koonpl ,
    I believe the model that you have converted expects FP16 values, which results in expecting 2x more bytes. So one way would be to convert UINT8 to FP16 by colorCamera.setFp16(True), and the other solution would be to add the conversion of UINT8->FP16 into the model itself. This can be done by specifying -ip U8 when using blobconverter:

    Thanks, Erik