HI,
I have trying to convent a propaietary from netowk to tidl.
my network has the below outputs:
Howver, the tidl adds the below layers(31 and on) aftewards(log from tidl_model_import.out)
27|TIDL_ConvolutionLayer |conv2d_9/Conv2D | 0| 1| 1| 26 x x x x x x x | 27 | 1 14 32 64 | 1 1 32 64 | 28672 |
28|TIDL_ConvolutionLayer |conv2d_10/Conv2D | 0| 1| 1| 26 x x x x x x x | 28 | 1 14 32 64 | 1 1 32 64 | 28672 |
29|TIDL_ConvolutionLayer |conv2d_11/Conv2D | 0| 1| 1| 26 x x x x x x x | 29 | 1 14 32 64 | 1 1 32 64 | 28672 |
30|TIDL_ConvolutionLayer |conv2d_12/Conv2D | 0| 1| 1| 26 x x x x x x x | 30 | 1 14 32 64 | 1 1 32 64 | 28672 |
31|TIDL_BatchNormLayer |lower_eyelid/Sigmoid | 0| 1| 1| 27 x x x x x x x | 31 | 1 1 32 64 | 1 1 32 64 | 2048 |
32|TIDL_BatchNormLayer |upper_eyelid/Sigmoid | 0| 1| 1| 28 x x x x x x x | 32 | 1 1 32 64 | 1 1 32 64 | 2048 |
33|TIDL_BatchNormLayer |pupil/Sigmoid | 0| 1| 1| 29 x x x x x x x | 33 | 1 1 32 64 | 1 1 32 64 | 2048 |
34|TIDL_BatchNormLayer |iris/Sigmoid | 0| 1| 1| 30 x x x x x x x | 34 | 1 1 32 64 | 1 1 32 64 | 2048 |
35|TIDL_DataLayer |lower_eyelid/Sigmoid | 0| 1| -1| 31 x x x x x x x | 0 | 1 1 32 64 | 0 0 0 0 | 0 |
36|TIDL_DataLayer |upper_eyelid/Sigmoid | 0| 1| -1| 32 x x x x x x x | 0 | 1 1 32 64 | 0 0 0 0 | 0 |
37|TIDL_DataLayer |pupil/Sigmoid | 0| 1| -1| 33 x x x x x x x | 0 | 1 1 32 64 | 0 0 0 0 | 0 |
38|TIDL_DataLayer |iris/Sigmoid | 0| 1| -1| 34 x x x x x x x | 0 | 1 1 32 64 | 0 0 0 0 | 0 |
---------------------------------------------
My conversion has two issue:
1. when runnig with trace, i see that the last layer that rertun meaningfull data is layer 30. from layer 31 all outputs are 0.
2. when running with the library ioBufDesc->outElementType[id] return 2 instreat of 6(float32)
for conversion i use: tidl_j7_01_03_00_11/ti_dl/utils/tidlModelImport/out/tidl_model_import.out with the below conigiration params:
modelType = 1
inputNetFile = "xxxx.pb"
outputNetFile = "xxxx.bin"
outputParamsFile = "xxxx.io"
outData = "/jacinto6/models/EyeHeatmapsNet_32x64_7Ch_for_ti_20220712T081657/out_tensord_out.txt"
inElementType = 0
foldBnInConv2D = 0
foldPreBnConv2D = 0
numParamBits = 16
numFeatureBits = 16
inDataNorm = 1
inMean = 128
inScale = 1
inWidth = 64
inHeight = 32
inNumChannels = 1
inFileFormat = 2
perfSimTool = /jacinto6/MPD_ENV/psdkra/tidl_j7_01_03_00_11/ti_dl/utils/perfsim/ti_cnnperfsim.out
perfSimConfig = /jacinto6/MPD_ENV/psdkra/tidl_j7_01_03_00_11/ti_dl/test/testvecs/config/import/device_config.cfg
inData = "/jacinto6/models/EyeHeatmapsNet_32x64_7Ch_for_ti_20220712T081657/images_list.txt"
postProcType = 0
tidlStatsTool = /jacinto6/MPD_ENV/psdkra/tidl_j7_01_03_00_11/ti_dl/test/PC_dsp_test_dl_algo.out
debugTraceLevel = 3
writeTraceLevel = 3
writeOutput = 2