TDA4VM: How to import a Unet model(onnx) for TDA4

Damon@INVO

Part Number: TDA4VM

Hi,

I'm trying to adopt tidl_ j7_ 08_ 01_ 00_ 0 tool to convert the Unet network model (onnx).The model is based on the Unet code on github（https://github.com/bubbliiiing/unet-pytorch).

When I inferenced on TIDL,the following erroe occurs:

I first convert the pytorch weight file to the onnx weight file,and tested the onnx weight file. The results of pytorch and onnx models are normal,but when inferencing and processing with .bin file,the above error occurred.

The cfg of import as follow:

modelType = 2
numParamBits = 16
numFeatureBits = 8
inputNetFile       = "../../test/testvecs/models/public/onnx/unet00.onnx"
outputNetFile      = "../../test/testvecs/config/tidl_models/onnx/tidl_net_unet.bin"
outputParamsFile   = "../../test/testvecs/config/tidl_models/onnx/tidl_net_unet_"
perfSimConfig = "../../test/testvecs/config/import/device_config.cfg"
postProcType = 3
inDataFormat = 0
inFileFormat = 2

inDataNorm = 1
inMean = 0 0 0
inScale = 0.003921 0.003921 0.003921

inWidth  = 320
inHeight = 320
inNumChannels = 3

inElementType = 0
quantizationStyle  = 3
inData  =   "../../test/testvecs/config/my_seg_list/NOR_19700101_012724_M_00069-00_2415.txt"