I am trying to run a non-image network through the NPU. I know you are meant to use the conversion scripts (ex 0_import, 1_quantize, 2_…) to convert a model in tensorflow .pb.
However, the parameters are image-specific as is the documentation
I am inputting a binary vector input – example shape (, 128). I don’t have channels, so the ordering and the quantizations and preprocessing assumed are off. And I am unsure how to handle the multiple inputs. Or what to do for the source file if I am not feeding in saved images…
Help setting params for 1d, non-image inputs/models…? I put my guesses, but they aren’t correct
This are my two tensorflow model architectures
$convert_tf
–tf-pb ./model/mymodel.pb
–inputs input
–input-size-list ‘,128’ ‘, 20’
–outputs maximum_1 activation_2
–net-output ${NAME}.json
–data-output ${NAME}.data
$tensorzone \
--action quantization \
--source text \
--source-file ./data/validation_tf.txt \
--channel-mean-value '128 0 0 1' \
--model-input ${NAME}.json \
--model-data ${NAME}.data \
--quantized-dtype asymmetric_quantized-u8 \
--quantized-rebuild
$tensorzone \
--action inference \
--source text \
--source-file ./data/validation_tf.txt \
--channel-mean-value '128 0 0 1' \
--model-input ${NAME}.json \
--model-data ${NAME}.data \
--dtype quantized
$export_ovxlib \
--model-input ${NAME}.json \
--data-input ${NAME}.data \
--reorder-channel '0 1 2' \
--channel-mean-value '128 0 0 1' \
--export-dtype quantized \
--model-quantize ${NAME}.quantize \
--optimize VIPNANOQI_PID0X88 \
--viv-sdk ../bin/vcmdtools \
--pack-nbg-unify