TDA4 model inference problem

Hi,

Do you have this file: tidl_j7_01_00_00_00\ti_dl\test\tidl_quant_stats_tool.out?

If yes then import did not happen successfully.

It could be due to unsupported layers or unsupported layer parameters.

For layer information you can refer to psdk_rtos_auto_j7_06_01_00_15/tidl_j7_01_00_00_00/ti_dl/docs/user_guide_html/md_tidl_layers_info.html

If you still face issues please share the import logs.

Regards,

Rishabh

Thanks for reply,

ConvolutionLayer ,DeconvLayer and ArgMaxLayer  are used in our model only.

After import I got the message as below:

~~~~~Running TIDL in PC emulation mode to collect Activations range for each layer~~~~~

Processing config file #0 : /home/tda4dev/psdk_rtos_auto_j7_06_01_00_15/tidl_j7_01_00_00_00/ti_dl/utils/tidlModelImport/tempDir/qunat_stats_config.txt
----------------------- TIDL Process with REF_ONLY FLOW ------------------------

# 0 . ..
------------------ Network Compiler Traces -----------------------------
Main iteration numer: 0....
Preparing for memory allocation : internal iteration number: 0
successful Memory allocation

-------------------- Network Compiler : Analysis Results are available --------------------

SUGGESTION: [TIDL_Deconv2DLayer] Panoptic3_up Please change to Upsample/Resize if possible. Upsample/Resize will be more efficient.
SUGGESTION: [TIDL_Deconv2DLayer] deconv2 Please change to Upsample/Resize if possible. Upsample/Resize will be more efficient.
SUGGESTION: [TIDL_Deconv2DLayer] deconv1 Please change to Upsample/Resize if possible. Upsample/Resize will be more efficient.
****************************************************
** ALL MODEL CHECK PASSED **
****************************************************

Thank you

Hi,

QS tool has crashed as there are no logs during import.

What are the kernel properties of de-convolution layers?

Please note that only 4x4 kernel with 2x2 stride is supported.

Regards,

Rishabh

Hi,

Deconv layer:

layer {
  name: "deconv1"
  type: "Deconvolution"
  bottom: "conv"
  top: "deconv1"
  param {
    lr_mult: 0.0
    decay_mult: 0.0
  }
  convolution_param {
    num_output: 8
    bias_term: false
    pad: 1
    kernel_size: 4
    group: 8
    stride: 2
    weight_filler {
      type: "bilinear"
    }
  }
}

This is the deconvolution setting.

Kernel_size = 4 and stride = 2 are the same in all deconv layer.

Thank you.