Hi Luxonis Team,

Is there a way to inspect quantized weights and layer properties after blob conversion? I have an ONNX regression model that, when converted, has significant error compared to the unconverted result.

Here are the my specifications on the online blobconverter tool:

Model Optimizer params: --data_type=FP16 --input_shape=[1,768,768,1]

Compile params: -ip FP16

I'd like to try "simulate" what's going on to understand if F32 to F16 quantization is the culprit, or if there is something else going on when running on the MyriadX. So far, I've tried to reproduce it in tensorflow using F16 quantized weights and that yields results much closer to the unconverted result.

    rsinghmn

    You can use numpy to load the .bin file, but not the .blob directly. Note that VPU layer implementations might differ slightly from those on-host, so this could cause the difference.

    Make sure that the input images you are feeding to the neural network are normalized in the same manner as well. This is one of the most common issues. By default, you will get BGR 0-255 images from camera node. If you model for example expects BGR 0-1, you need to also provide --scale 255 and --reverse_input_channels to model optimizer.