Hi,
Tried converting classifier models such as InceptionV3, Resnet50 and Mobilenetv3Small using neutron converter targeting iMX95. On converting fully connected layer to ConvertFullyConnectedToConv, it is showing internal error.
This is the fully connected layer which is not getting converted. But this layer does not violate any of the constraints mentioned in the section 5.9.7 FULLY_CONNECTED constraints of UG.
How can we fix this issue as I am unable to run any classifier with fully connected layer in iMX95 Neutron NPU.
Hi Pooja,
I am also facing this issue that fully_connected layer not able to convert into neutron graph.
Did you find any solution of it?
Regards,
Varshil
i.mx95 is still under pre-prodcution, we couldn't share detailed technical information on the public community, pls contact local DFAE to get further support