TDA4VM: Edgeai: Calibration issue with Batch Normalization layer implemented for constant multiply broadcast.

Deepanshu Yadvandu
Part Number: TDA4VM
Tool/software:
Hi,
We are trying to run out model on EdgeAI 9.20.07 https://github.com/TexasInstruments/edgeai-tidl-tools/tree/09_02_07_00?tab=readme-ov-file.
There is a problem with the int16/int8 quantization calibration when multiply broadcast with a constant is implemented with Batch Normalization as is done in SDK9.2. The first quantized calibration iteration runs fine, but in the second iteration the multiply broadcast layer returns all zeros and in the third iteration it returns all -1s.
Please Note that the above problem doesn't occur when the constant multiply broadcast is implemented as element wise layer.
7457.4_fnc_int16.txt
over 1 year ago
0 Pratik Kedar over 1 year ago
TI__Mastermind 24041 points
Please validate the observation on latest tidl tools SDK tag : 9.2.9.0
0 Deepanshu Yadvandu over 1 year ago in reply to Pratik Kedar
Prodigy 90 points
EdgeAI default tidl_tools have never worked for us, EdgeAI 9.2.9 https://github.com/TexasInstruments/edgeai-tidl-tools/tree/09_02_09_00 also doesn't work for us. It gives error, free(): double free detected in tcache 2.
Fullscreen log_fp32.txt Download
***** WARNING : tensor_bits = 32 -- Compiling for floating point - target execution is not supported for 32 bit compilation !! ***** 
tidl_tools_path                                 = /home/deepanshu/EdgeAI/new/edgeai-tidl-tools/tidl_tools 
artifacts_folder                                = ../../../model-artifacts//fnc_safety_fp32/ 
tidl_tensor_bits                                = 32 
debug_level                                     = 4 
num_tidl_subgraphs                              = 16 
tidl_denylist                                   = 
tidl_denylist_layer_name                        =    
tidl_denylist_layer_type                         = 
tidl_allowlist_layer_name                        = 
model_type                                      =  
tidl_calibration_accuracy_level                 = 7 
tidl_calibration_options:num_frames_calibration = 3 
tidl_calibration_options:bias_calibration_iterations = 3 
mixed_precision_factor = -1.000000 
model_group_id = 0 
power_of_2_quantization                         = 2 
ONNX QDQ Enabled                                = 0 
enable_high_resolution_optimization             = 0 
pre_batchnorm_fold                              = 1 
add_data_convert_ops                            = 3 
output_feature_16bit_names_list                 =  
m_params_16bit_names_list                       =  
m_single_core_layers_names_list                    =  
reserved_compile_constraints_flag               = 1601 
ti_internal_reserved_1                          = 


WARNING : 'meta_layers_names_list' is not provided - running OD post processing in ARM mode 
 
Number of OD backbone nodes = 0 
Size of odBackboneNodeIds = 0 
Layer 0 -- layer name -- Conv_0 
 Input dims size = 4     dims --- 1   3   320   800   
Supported TIDL layer type ---            Conv -- Conv_0 
Layer 1 -- layer name -- Relu_1 
 Input dims size = 4     dims --- 1   64   160   400   
Supported TIDL layer type ---            Relu -- Relu_1 
Layer 2 -- layer name -- MaxPool_2 
 Input dims size = 4     dims --- 1   64   160   400   
Supported TIDL layer type ---         MaxPool -- MaxPool_2 
Layer 3 -- layer name -- Conv_3 
 Input dims size = 4     dims --- 1   64   80   200   
Supported TIDL layer type ---            Conv -- Conv_3 
Layer 4 -- layer name -- Relu_4 
 Input dims size = 4     dims --- 1   64   80   200   
Supported TIDL layer type ---            Relu -- Relu_4 
Layer 5 -- layer name -- Conv_5 
 Input dims size = 4     dims --- 1   64   80   200   
Supported TIDL layer type ---            Conv -- Conv_5 
Layer 6 -- layer name -- Add_6 
 Input dims size = 4     dims --- 1   64   80   200   
Supported TIDL layer type ---             Add -- Add_6 
Layer 7 -- layer name -- Relu_7 
 Input dims size = 4     dims --- 1   64   80   200   
Supported TIDL layer type ---            Relu -- Relu_7 
Layer 8 -- layer name -- Conv_8 
 Input dims size = 4     dims --- 1   64   80   200   
Supported TIDL layer type ---            Conv -- Conv_8 
Layer 9 -- layer name -- Relu_9 
 Input dims size = 4     dims --- 1   64   80   200   
Supported TIDL layer type ---            Relu -- Relu_9 
Layer 10 -- layer name -- Conv_10 
 Input dims size = 4     dims --- 1   64   80   200   
Supported TIDL layer type ---            Conv -- Conv_10 
Layer 11 -- layer name -- Add_11 
 Input dims size = 4     dims --- 1   64   80   200   
Supported TIDL layer type ---             Add -- Add_11 
Layer 12 -- layer name -- Relu_12 
 Input dims size = 4     dims --- 1   64   80   200   
Supported TIDL layer type ---            Relu -- Relu_12 
Layer 13 -- layer name -- Conv_13 
 Input dims size = 4     dims --- 1   64   80   200   
Supported TIDL layer type ---            Conv -- Conv_13 
Layer 14 -- layer name -- Relu_14 
 Input dims size = 4     dims --- 1   128   40   100   
Supported TIDL layer type ---            Relu -- Relu_14 
Layer 15 -- layer name -- Conv_15 
 Input dims size = 4     dims --- 1   128   40   100   
Supported TIDL layer type ---            Conv -- Conv_15 
Layer 16 -- layer name -- Conv_16 
 Input dims size = 4     dims --- 1   64   80   200   
Supported TIDL layer type ---            Conv -- Conv_16 
Layer 17 -- layer name -- Add_17 
 Input dims size = 4     dims --- 1   128   40   100   
Supported TIDL layer type ---             Add -- Add_17 
Layer 18 -- layer name -- Relu_18 
 Input dims size = 4     dims --- 1   128   40   100   
Supported TIDL layer type ---            Relu -- Relu_18 
Layer 19 -- layer name -- Conv_19 
 Input dims size = 4     dims --- 1   128   40   100   
Supported TIDL layer type ---            Conv -- Conv_19 
Layer 20 -- layer name -- Relu_20 
 Input dims size = 4     dims --- 1   128   40   100   
Supported TIDL layer type ---            Relu -- Relu_20 
Layer 21 -- layer name -- Conv_21 
 Input dims size = 4     dims --- 1   128   40   100   
Supported TIDL layer type ---            Conv -- Conv_21 
Layer 22 -- layer name -- Add_22 
 Input dims size = 4     dims --- 1   128   40   100   
Supported TIDL layer type ---             Add -- Add_22 
Layer 23 -- layer name -- Relu_23 
 Input dims size = 4     dims --- 1   128   40   100   
Supported TIDL layer type ---            Relu -- Relu_23 
Layer 24 -- layer name -- Conv_24 
 Input dims size = 4     dims --- 1   128   40   100   
Supported TIDL layer type ---            Conv -- Conv_24 
Layer 25 -- layer name -- Relu_25 
 Input dims size = 4     dims --- 1   256   20   50   
Supported TIDL layer type ---            Relu -- Relu_25 
Layer 26 -- layer name -- Conv_26 
 Input dims size = 4     dims --- 1   256   20   50   
Supported TIDL layer type ---            Conv -- Conv_26 
Layer 27 -- layer name -- Conv_27 
 Input dims size = 4     dims --- 1   128   40   100   
Supported TIDL layer type ---            Conv -- Conv_27 
Layer 28 -- layer name -- Add_28 
 Input dims size = 4     dims --- 1   256   20   50   
Supported TIDL layer type ---             Add -- Add_28 
Layer 29 -- layer name -- Relu_29 
 Input dims size = 4     dims --- 1   256   20   50   
Supported TIDL layer type ---            Relu -- Relu_29 
Layer 30 -- layer name -- Conv_30 
 Input dims size = 4     dims --- 1   256   20   50   
Supported TIDL layer type ---            Conv -- Conv_30 
Layer 31 -- layer name -- Relu_31 
 Input dims size = 4     dims --- 1   256   20   50   
Supported TIDL layer type ---            Relu -- Relu_31 
Layer 32 -- layer name -- Conv_32 
 Input dims size = 4     dims --- 1   256   20   50   
Supported TIDL layer type ---            Conv -- Conv_32 
Layer 33 -- layer name -- Add_33 
 Input dims size = 4     dims --- 1   256   20   50   
Supported TIDL layer type ---             Add -- Add_33 
Layer 34 -- layer name -- Relu_34 
 Input dims size = 4     dims --- 1   256   20   50   
Supported TIDL layer type ---            Relu -- Relu_34 
Layer 35 -- layer name -- Conv_35 
 Input dims size = 4     dims --- 1   256   20   50   
Supported TIDL layer type ---            Conv -- Conv_35 
Layer 36 -- layer name -- Relu_36 
 Input dims size = 4     dims --- 1   512   10   25   
Supported TIDL layer type ---            Relu -- Relu_36 
Layer 37 -- layer name -- Conv_37 
 Input dims size = 4     dims --- 1   512   10   25   
Supported TIDL layer type ---            Conv -- Conv_37 
Layer 38 -- layer name -- Conv_38 
 Input dims size = 4     dims --- 1   256   20   50   
Supported TIDL layer type ---            Conv -- Conv_38 
Layer 39 -- layer name -- Add_39 
 Input dims size = 4     dims --- 1   512   10   25   
Supported TIDL layer type ---             Add -- Add_39 
Layer 40 -- layer name -- Relu_40 
 Input dims size = 4     dims --- 1   512   10   25   
Supported TIDL layer type ---            Relu -- Relu_40 
Layer 41 -- layer name -- Conv_41 
 Input dims size = 4     dims --- 1   512   10   25   
Supported TIDL layer type ---            Conv -- Conv_41 
Layer 42 -- layer name -- Relu_42 
 Input dims size = 4     dims --- 1   512   10   25   
Supported TIDL layer type ---            Relu -- Relu_42 
Layer 43 -- layer name -- Conv_43 
 Input dims size = 4     dims --- 1   512   10   25   
Supported TIDL layer type ---            Conv -- Conv_43 
Layer 44 -- layer name -- Add_44 
 Input dims size = 4     dims --- 1   512   10   25   
Supported TIDL layer type ---             Add -- Add_44 
Layer 45 -- layer name -- Relu_45 
 Input dims size = 4     dims --- 1   512   10   25   
Supported TIDL layer type ---            Relu -- Relu_45 
Layer 46 -- layer name -- Conv_46 
 Input dims size = 4     dims --- 1   512   10   25   
Supported TIDL layer type ---            Conv -- Conv_46 
Layer 47 -- layer name -- Relu_47 
 Input dims size = 4     dims --- 1   64   10   25   
Supported TIDL layer type ---            Relu -- Relu_47 
Layer 48 -- layer name -- Add_58 
 Input dims size = 4     dims --- 1   64   10   25   
Supported TIDL layer type ---             Add -- Add_58 
Layer 49 -- layer name -- Conv_59 
 Input dims size = 4     dims --- 1   64   10   25   
Supported TIDL layer type ---            Conv -- Conv_59 
Layer 50 -- layer name -- Reshape_64 
 Input dims size = 4     dims --- 1   16   10   25   
Supported TIDL layer type ---         Reshape -- Reshape_64 
Layer 51 -- layer name -- Transpose_65 
 Input dims size = 3     dims --- 1   16   250   
Supported TIDL layer type ---       Transpose -- Transpose_65 
Layer 52 -- layer name -- Conv_66 
 Input dims size = 4     dims --- 1   64   10   25   
Supported TIDL layer type ---            Conv -- Conv_66 
Layer 53 -- layer name -- Reshape_71 
 Input dims size = 4     dims --- 1   16   10   25   
Supported TIDL layer type ---         Reshape -- Reshape_71 
Layer 54 -- layer name -- MatMul_72 
 Input dims size = 3     dims --- 1   250   16   
Supported TIDL layer type ---          MatMul -- MatMul_72 
Layer 55 -- layer name -- reshape_Unsqueeze_73 
 Input dims size = 3     dims --- 1   250   250   
Supported TIDL layer type ---         Reshape -- reshape_Unsqueeze_73 
Layer 56 -- layer name -- Softmax_74 
 Input dims size = 4     dims --- 1   1   250   250   
Supported TIDL layer type ---         Softmax -- Softmax_74 
Layer 57 -- layer name -- Squeeze_75 
 Input dims size = 4     dims --- 1   1   250   250   
Supported TIDL layer type ---         Squeeze -- Squeeze_75 
Layer 58 -- layer name -- Transpose_76 
 Input dims size = 3     dims --- 1   250   250   
Supported TIDL layer type ---       Transpose -- Transpose_76 
Layer 59 -- layer name -- Conv_77 
 Input dims size = 4     dims --- 1   64   10   25   
Supported TIDL layer type ---            Conv -- Conv_77 
Layer 60 -- layer name -- Reshape_82 
 Input dims size = 4     dims --- 1   64   10   25   
Supported TIDL layer type ---         Reshape -- Reshape_82 
Layer 61 -- layer name -- MatMul_83 
 Input dims size = 3     dims --- 1   64   250   
Supported TIDL layer type ---          MatMul -- MatMul_83 
Layer 62 -- layer name -- Reshape_88 
 Input dims size = 3     dims --- 1   64   250   
Supported TIDL layer type ---         Reshape -- Reshape_88 
Layer 63 -- layer name -- Mul_89 
 Input dims size = 0     dims --- 
Supported TIDL layer type ---             Mul -- Mul_89 
Layer 64 -- layer name -- Add_90 
 Input dims size = 4     dims --- 1   64   10   25   
Supported TIDL layer type ---             Add -- Add_90 
Layer 65 -- layer name -- Conv_91 
 Input dims size = 4     dims --- 1   64   10   25   
Supported TIDL layer type ---            Conv -- Conv_91 
Layer 66 -- layer name -- Relu_92 
 Input dims size = 4     dims --- 1   64   10   25   
Supported TIDL layer type ---            Relu -- Relu_92 
Layer 67 -- layer name -- Conv_93 
 Input dims size = 4     dims --- 1   64   10   25   
Supported TIDL layer type ---            Conv -- Conv_93 
Layer 68 -- layer name -- Relu_94 
 Input dims size = 4     dims --- 1   64   10   25   
Supported TIDL layer type ---            Relu -- Relu_94 
Layer 69 -- layer name -- Add_105 
 Input dims size = 4     dims --- 1   64   10   25   
Supported TIDL layer type ---             Add -- Add_105 
Layer 70 -- layer name -- Conv_106 
 Input dims size = 4     dims --- 1   64   10   25   
Supported TIDL layer type ---            Conv -- Conv_106 
Layer 71 -- layer name -- Reshape_111 
 Input dims size = 4     dims --- 1   16   10   25   
Supported TIDL layer type ---         Reshape -- Reshape_111 
Layer 72 -- layer name -- Transpose_112 
 Input dims size = 3     dims --- 1   16   250   
Supported TIDL layer type ---       Transpose -- Transpose_112 
Layer 73 -- layer name -- Conv_113 
 Input dims size = 4     dims --- 1   64   10   25   
Supported TIDL layer type ---            Conv -- Conv_113 
Layer 74 -- layer name -- Reshape_118 
 Input dims size = 4     dims --- 1   16   10   25   
Supported TIDL layer type ---         Reshape -- Reshape_118 
Layer 75 -- layer name -- MatMul_119 
 Input dims size = 3     dims --- 1   250   16   
Supported TIDL layer type ---          MatMul -- MatMul_119 
Layer 76 -- layer name -- reshape_Unsqueeze_120 
 Input dims size = 3     dims --- 1   250   250   
Supported TIDL layer type ---         Reshape -- reshape_Unsqueeze_120 
Layer 77 -- layer name -- Softmax_121 
 Input dims size = 4     dims --- 1   1   250   250   
Supported TIDL layer type ---         Softmax -- Softmax_121 
Layer 78 -- layer name -- Squeeze_122 
 Input dims size = 4     dims --- 1   1   250   250   
Supported TIDL layer type ---         Squeeze -- Squeeze_122 
Layer 79 -- layer name -- Transpose_123 
 Input dims size = 3     dims --- 1   250   250   
Supported TIDL layer type ---       Transpose -- Transpose_123 
Layer 80 -- layer name -- Conv_124 
 Input dims size = 4     dims --- 1   64   10   25   
Supported TIDL layer type ---            Conv -- Conv_124 
Layer 81 -- layer name -- Reshape_129 
 Input dims size = 4     dims --- 1   64   10   25   
Supported TIDL layer type ---         Reshape -- Reshape_129 
Layer 82 -- layer name -- MatMul_130 
 Input dims size = 3     dims --- 1   64   250   
Supported TIDL layer type ---          MatMul -- MatMul_130 
Layer 83 -- layer name -- Reshape_135 
 Input dims size = 3     dims --- 1   64   250   
Supported TIDL layer type ---         Reshape -- Reshape_135 
Layer 84 -- layer name -- Mul_136 
 Input dims size = 0     dims --- 
Supported TIDL layer type ---             Mul -- Mul_136 
Layer 85 -- layer name -- Add_137 
 Input dims size = 4     dims --- 1   64   10   25   
Supported TIDL layer type ---             Add -- Add_137 
Layer 86 -- layer name -- Conv_138 
 Input dims size = 4     dims --- 1   64   10   25   
Supported TIDL layer type ---            Conv -- Conv_138 
Layer 87 -- layer name -- Relu_139 
 Input dims size = 4     dims --- 1   64   10   25   
Supported TIDL layer type ---            Relu -- Relu_139 
Layer 88 -- layer name -- Conv_142 
 Input dims size = 4     dims --- 1   64   10   25   
Supported TIDL layer type ---            Conv -- Conv_142 
Layer 89 -- layer name -- Upsample_143 
 Input dims size = 4     dims --- 1   64   10   25   
Supported TIDL layer type ---        Upsample -- Upsample_143 
Layer 90 -- layer name -- Conv_141 
 Input dims size = 4     dims --- 1   256   20   50   
Supported TIDL layer type ---            Conv -- Conv_141 
Layer 91 -- layer name -- Add_144 
 Input dims size = 4     dims --- 1   64   20   50   
Supported TIDL layer type ---             Add -- Add_144 
Layer 92 -- layer name -- Conv_148 
 Input dims size = 4     dims --- 1   64   20   50   
Supported TIDL layer type ---            Conv -- Conv_148 
Layer 93 -- layer name -- Conv_152 
 Input dims size = 4     dims --- 1   64   20   50   
Supported TIDL layer type ---            Conv -- Conv_152 
Layer 94 -- layer name -- Relu_153 
 Input dims size = 4     dims --- 1   64   20   50   
Supported TIDL layer type ---            Relu -- Relu_153 
Layer 95 -- layer name -- Conv_154 
 Input dims size = 4     dims --- 1   64   20   50   
Supported TIDL layer type ---            Conv -- Conv_154 
Layer 96 -- layer name -- Sigmoid_170 
 Input dims size = 4     dims --- 1   1   20   50   
Supported TIDL layer type ---         Sigmoid -- Sigmoid_170 
Layer 97 -- layer name -- Add_172 
 Input dims size = 4     dims --- 1   1   20   50   
Supported TIDL layer type ---             Add -- Add_172 
Layer 98 -- layer name -- Relu_173 
 Input dims size = 4     dims --- 1   1   20   50   
Supported TIDL layer type ---            Relu -- Relu_173 
Layer 99 -- layer name -- Add_175 
 Input dims size = 4     dims --- 1   1   20   50   
Supported TIDL layer type ---             Add -- Add_175 
Layer 100 -- layer name -- Mul_177 
 Input dims size = 0     dims --- 
Supported TIDL layer type ---             Mul -- Mul_177 
Layer 101 -- layer name -- Add_179 
 Input dims size = 0     dims --- 
Supported TIDL layer type ---             Add -- Add_179 
Layer 102 -- layer name -- Relu_180 
 Input dims size = 4     dims --- 1   1   20   50   
Supported TIDL layer type ---            Relu -- Relu_180 
Layer 103 -- layer name -- Mul_182 
 Input dims size = 0     dims --- 
Supported TIDL layer type ---             Mul -- Mul_182 
Layer 104 -- layer name -- Add_184 
 Input dims size = 0     dims --- 
Supported TIDL layer type ---             Add -- Add_184 
Layer 105 -- layer name -- MaxPool_204 
 Input dims size = 4     dims --- 1   1   20   50   
Supported TIDL layer type ---         MaxPool -- MaxPool_204 
Layer 106 -- layer name -- Mul_206 
 Input dims size = 4     dims --- 1   1   20   50   
Supported TIDL layer type ---             Mul -- Mul_206 
Layer 107 -- layer name -- Add_207 
 Input dims size = 4     dims --- 1   1   20   50   
Supported TIDL layer type ---             Add -- Add_207 
Layer 108 -- layer name -- Mul_209 
 Input dims size = 4     dims --- 1   1   20   50   
Supported TIDL layer type ---             Mul -- Mul_209 
Layer 109 -- layer name -- Add_211 
 Input dims size = 4     dims --- 1   1   20   50   
Supported TIDL layer type ---             Add -- Add_211 
Layer 110 -- layer name -- Relu_212 
 Input dims size = 4     dims --- 1   1   20   50   
Supported TIDL layer type ---            Relu -- Relu_212 
Layer 111 -- layer name -- Mul_214 
 Input dims size = 4     dims --- 1   1   20   50   
Supported TIDL layer type ---             Mul -- Mul_214 
Layer 112 -- layer name -- Mul_215 
 Input dims size = 4     dims --- 1   1   20   50   
Supported TIDL layer type ---             Mul -- Mul_215 
Layer 113 -- layer name -- Squeeze_216 
 Input dims size = 4     dims --- 1   1   20   50   
Supported TIDL layer type ---         Squeeze -- Squeeze_216 
Layer 114 -- layer name -- Reshape_218 
 Input dims size = 3     dims --- 1   20   50   
Supported TIDL layer type ---         Reshape -- Reshape_218 
Layer 115 -- layer name -- TopK_219 
 Input dims size = 1     dims --- 1000   
Unsupported (imports) TIDL layer type for ONNX op type --- TopK 
Layer 116 -- layer name -- Reshape_221 
 Input dims size = 1     dims --- 6   
Supported TIDL layer type ---         Reshape -- Reshape_221 
Layer 117 -- layer name -- Squeeze_224 
 Input dims size = 2     dims --- 6   1   
Supported TIDL layer type ---         Squeeze -- Squeeze_224 
Layer 118 -- layer name -- Conv_155 
 Input dims size = 4     dims --- 1   64   20   50   
Supported TIDL layer type ---            Conv -- Conv_155 
Layer 119 -- layer name -- Relu_156 
 Input dims size = 4     dims --- 1   64   20   50   
Supported TIDL layer type ---            Relu -- Relu_156 
Layer 120 -- layer name -- Conv_157 
 Input dims size = 4     dims --- 1   64   20   50   
Supported TIDL layer type ---            Conv -- Conv_157 
Layer 121 -- layer name -- Reshape_195 
 Input dims size = 4     dims --- 1   134   20   50   
Supported TIDL layer type ---         Reshape -- Reshape_195 
Layer 122 -- layer name -- Transpose_196 
 Input dims size = 5     dims --- 1   1   134   20   50   
Supported TIDL layer type ---       Transpose -- Transpose_196 
Layer 123 -- layer name -- Reshape_203 
 Input dims size = 5     dims --- 1   134   1   20   50   
Supported TIDL layer type ---         Reshape -- Reshape_203 
Layer 124 -- layer name -- Conv_228 
 Input dims size = 4     dims --- 1   134   1   1000   
Supported TIDL layer type ---            Conv -- Conv_228 
Layer 125 -- layer name -- Transpose_229 
 Input dims size = 4     dims --- 1   67   1   1000   
Supported TIDL layer type ---       Transpose -- Transpose_229 
Layer 126 -- layer name -- Reshape_231 
 Input dims size = 4     dims --- 1   1000   1   67   
Supported TIDL layer type ---         Reshape -- Reshape_231 
Layer 127 -- layer name -- Gather_233 
 Input dims size = 2     dims --- 1000   67   
Supported TIDL layer type ---          Gather -- Gather_233 
Layer 128 -- layer name -- Transpose_245 
 Input dims size = 2     dims --- 6   67   
Supported TIDL layer type ---       Transpose -- Transpose_245 
Layer 129 -- layer name -- Reshape_248 
 Input dims size = 2     dims --- 67   6   
Supported TIDL layer type ---         Reshape -- Reshape_248 
Layer 130 -- layer name -- Conv_253 
 Input dims size = 4     dims --- 1   67   1   6   
Supported TIDL layer type ---            Conv -- Conv_253 
Layer 131 -- layer name -- Transpose_254 
 Input dims size = 4     dims --- 1   1   1   6   
Supported TIDL layer type ---       Transpose -- Transpose_254 
Layer 132 -- layer name -- Reshape_256 
 Input dims size = 4     dims --- 1   6   1   1   
Supported TIDL layer type ---         Reshape -- Reshape_256 
Layer 133 -- layer name -- Add_263 
 Input dims size = 2     dims --- 6   1   
Supported TIDL layer type ---             Add -- Add_263 
Layer 134 -- layer name -- Reshape_273 
 Input dims size = 2     dims --- 6   1   
Supported TIDL layer type ---         Reshape -- Reshape_273 
Layer 135 -- layer name -- Reshape_283 
 Input dims size = 1     dims --- 6   
Supported TIDL layer type ---         Reshape -- Reshape_283 
Layer 136 -- layer name -- Upsample_145 
 Input dims size = 4     dims --- 1   64   20   50   
Supported TIDL layer type ---        Upsample -- Upsample_145 
Layer 137 -- layer name -- Conv_140 
 Input dims size = 4     dims --- 1   128   40   100   
Supported TIDL layer type ---            Conv -- Conv_140 
Layer 138 -- layer name -- Add_146 
 Input dims size = 4     dims --- 1   64   40   100   
Supported TIDL layer type ---             Add -- Add_146 
Layer 139 -- layer name -- Conv_147 
 Input dims size = 4     dims --- 1   64   40   100   
Supported TIDL layer type ---            Conv -- Conv_147 
Layer 140 -- layer name -- Conv_185 
 Input dims size = 4     dims --- 1   64   40   100   
Supported TIDL layer type ---            Conv -- Conv_185 
Layer 141 -- layer name -- Relu_186 
 Input dims size = 4     dims --- 1   64   40   100   
Supported TIDL layer type ---            Relu -- Relu_186 
Layer 142 -- layer name -- Conv_187 
 Input dims size = 4     dims --- 1   64   40   100   
Supported TIDL layer type ---            Conv -- Conv_187 
Layer 143 -- layer name -- Relu_188 
 Input dims size = 4     dims --- 1   64   40   100   
Supported TIDL layer type ---            Relu -- Relu_188 
Layer 144 -- layer name -- Conv_189 
 Input dims size = 4     dims --- 1   64   40   100   
Supported TIDL layer type ---            Conv -- Conv_189 
Layer 145 -- layer name -- Relu_190 
 Input dims size = 4     dims --- 1   64   40   100   
Supported TIDL layer type ---            Relu -- Relu_190 
Layer 146 -- layer name -- Concat_241 
 Input dims size = 0     dims --- 
Supported TIDL layer type ---          Concat -- Concat_241 
Layer 147 -- layer name -- Reshape_rc_0 
 Input dims size = 4     dims --- 1   66   40   100   
Supported TIDL layer type ---         Reshape -- Reshape_rc_0 
Layer 148 -- layer name -- Conv_249 
 Input dims size = 4     dims --- 1   67   1   6   
Supported TIDL layer type ---            Conv -- Conv_249 
Layer 149 -- layer name -- Transpose_250 
 Input dims size = 4     dims --- 1   66   1   6   
Supported TIDL layer type ---       Transpose -- Transpose_250 
Layer 150 -- layer name -- Reshape_252 
 Input dims size = 4     dims --- 1   6   1   66   
Supported TIDL layer type ---         Reshape -- Reshape_252 
Layer 151 -- layer name -- MatMul_rc_0 
 Input dims size = 2     dims --- 6   66   
Supported TIDL layer type ---          MatMul -- MatMul_rc_0 
Layer 152 -- layer name -- Reshape_rc_1 
 Input dims size = 2     dims --- 6   4000   
Supported TIDL layer type ---         Reshape -- Reshape_rc_1 
Layer 153 -- layer name -- Add_284 
 Input dims size = 4     dims --- 1   6   40   100   
Unsupported (TIDL check) TIDL layer type ---             Add 
Layer 154 -- layer name -- Reshape_288 
 Input dims size = 4     dims --- 1   6   40   100   
Supported TIDL layer type ---         Reshape -- Reshape_288 
Layer 155 -- layer name -- Transpose_n_1 
 Input dims size = 4     dims --- 1   6   40   100   
Supported TIDL layer type ---       Transpose -- Transpose_n_1 
Layer 156 -- layer name -- Conv3_2 
 Input dims size = 4     dims --- 1   100   6   40   
Supported TIDL layer type ---            Conv -- Conv3_2 
Layer 157 -- layer name -- Relu_359 
 Input dims size = 4     dims --- 1   64   6   40   
Supported TIDL layer type ---            Relu -- Relu_359 
Layer 158 -- layer name -- Conv3_4 
 Input dims size = 4     dims --- 1   64   6   40   
Supported TIDL layer type ---            Conv -- Conv3_4 
Layer 159 -- layer name -- Transpose_n_6 
 Input dims size = 4     dims --- 1   2   6   40   
Supported TIDL layer type ---       Transpose -- Transpose_n_6 
Layer 160 -- layer name -- Conv_363 
 Input dims size = 4     dims --- 1   100   6   40   
Supported TIDL layer type ---            Conv -- Conv_363 
Layer 161 -- layer name -- Relu_364 
 Input dims size = 4     dims --- 1   1   6   40   
Supported TIDL layer type ---            Relu -- Relu_364 
Layer 162 -- layer name -- Transpose_365 
 Input dims size = 4     dims --- 1   1   6   40   
Supported TIDL layer type ---       Transpose -- Transpose_365 
Layer 163 -- layer name -- Conv_366 
 Input dims size = 4     dims --- 1   40   1   6   
Supported TIDL layer type ---            Conv -- Conv_366 
Layer 164 -- layer name -- Transpose_367 
 Input dims size = 4     dims --- 1   8   1   6   
Supported TIDL layer type ---       Transpose -- Transpose_367 
Layer 165 -- layer name -- Conv_289 
 Input dims size = 4     dims --- 1   134   1   1000   
Supported TIDL layer type ---            Conv -- Conv_289 
Layer 166 -- layer name -- Transpose_290 
 Input dims size = 4     dims --- 1   67   1   1000   
Supported TIDL layer type ---       Transpose -- Transpose_290 
Layer 167 -- layer name -- Reshape_292 
 Input dims size = 4     dims --- 1   1000   1   67   
Supported TIDL layer type ---         Reshape -- Reshape_292 
Layer 168 -- layer name -- Gather_294 
 Input dims size = 2     dims --- 1000   67   
Supported TIDL layer type ---          Gather -- Gather_294 
Layer 169 -- layer name -- Transpose_306 
 Input dims size = 2     dims --- 6   67   
Supported TIDL layer type ---       Transpose -- Transpose_306 
Layer 170 -- layer name -- Reshape_309 
 Input dims size = 2     dims --- 67   6   
Supported TIDL layer type ---         Reshape -- Reshape_309 
Layer 171 -- layer name -- Conv_314 
 Input dims size = 4     dims --- 1   67   1   6   
Supported TIDL layer type ---            Conv -- Conv_314 
Layer 172 -- layer name -- Transpose_315 
 Input dims size = 4     dims --- 1   1   1   6   
Supported TIDL layer type ---       Transpose -- Transpose_315 
Layer 173 -- layer name -- Reshape_317 
 Input dims size = 4     dims --- 1   6   1   1   
Supported TIDL layer type ---         Reshape -- Reshape_317 
Layer 174 -- layer name -- Add_324 
 Input dims size = 2     dims --- 6   1   
Supported TIDL layer type ---             Add -- Add_324 
Layer 175 -- layer name -- Reshape_334 
 Input dims size = 2     dims --- 6   1   
Supported TIDL layer type ---         Reshape -- Reshape_334 
Layer 176 -- layer name -- Reshape_344 
 Input dims size = 1     dims --- 6   
Supported TIDL layer type ---         Reshape -- Reshape_344 
Layer 177 -- layer name -- Conv_310 
 Input dims size = 4     dims --- 1   67   1   6   
Supported TIDL layer type ---            Conv -- Conv_310 
Layer 178 -- layer name -- Transpose_311 
 Input dims size = 4     dims --- 1   66   1   6   
Supported TIDL layer type ---       Transpose -- Transpose_311 
Layer 179 -- layer name -- Reshape_313 
 Input dims size = 4     dims --- 1   6   1   66   
Supported TIDL layer type ---         Reshape -- Reshape_313 
Layer 180 -- layer name -- MatMul_rc_1 
 Input dims size = 2     dims --- 6   66   
Supported TIDL layer type ---          MatMul -- MatMul_rc_1 
Layer 181 -- layer name -- Reshape_rc_2 
 Input dims size = 2     dims --- 6   4000   
Supported TIDL layer type ---         Reshape -- Reshape_rc_2 
Layer 182 -- layer name -- Add_345 
 Input dims size = 4     dims --- 1   6   40   100   
Unsupported (TIDL check) TIDL layer type ---             Add 
Layer 183 -- layer name -- Reshape_349 
 Input dims size = 4     dims --- 1   6   40   100   
Supported TIDL layer type ---         Reshape -- Reshape_349 

Preliminary subgraphs created = 4 
Final number of subgraphs created are : 4, - Offloaded Nodes - 181, Total Nodes - 184 
INFORMATION -- [TIDL_ResizeLayer]  Any resize ratio which is power of 2 and greater than 4 will be placed by combination of 4x4 resize layer and 2x2 resize layer. For example a 8x8 resize will be replaced by 4x4 resize followed by 2x2 resize.  
Layers type not supported by TIDL --- layer type - TopK,  Node name -TopK_219  
INFORMATION -- [TIDL_ResizeLayer]  Any resize ratio which is power of 2 and greater than 4 will be placed by combination of 4x4 resize layer and 2x2 resize layer. For example a 8x8 resize will be replaced by 4x4 resize followed by 2x2 resize.  
ALLOWLISTING : ADD/MUL/SUB/DIV layer : The variable inputs must of be same dimensions --tidl_import_common_model_check.cpp , TIDL_checkAddMulSubDivTensorProperties , 403  
ALLOWLISTING : ADD/MUL/SUB/DIV layer : The variable inputs must of be same dimensions --tidl_import_common_model_check.cpp , TIDL_checkAddMulSubDivTensorProperties , 403  
Running runtimes graphviz - /home/deepanshu/EdgeAI/new/edgeai-tidl-tools/tidl_tools/tidl_graphVisualiser_runtimes.out ../../../model-artifacts//fnc_safety_fp32//allowedNode.txt ../../../model-artifacts//fnc_safety_fp32//tempDir/graphvizInfo.txt ../../../model-artifacts//fnc_safety_fp32//tempDir/runtimes_visualization.svg 
*** In TIDL_createStateImportFunc *** 
Compute on node : TIDLExecutionProvider_TIDL_0_0
  0,            Conv, 3, 1, image, 717
  1,            Relu, 1, 1, 717, 227
  2,         MaxPool, 1, 1, 227, 228
  3,            Conv, 3, 1, 228, 720
  4,            Relu, 1, 1, 720, 231
  5,            Conv, 3, 1, 231, 723
  6,             Add, 2, 1, 723, 234
  7,            Relu, 1, 1, 234, 235
  8,            Conv, 3, 1, 235, 726
  9,            Relu, 1, 1, 726, 238
 10,            Conv, 3, 1, 238, 729
 11,             Add, 2, 1, 729, 241
 12,            Relu, 1, 1, 241, 242
 13,            Conv, 3, 1, 242, 738
 14,            Conv, 3, 1, 242, 732
 15,            Relu, 1, 1, 732, 245
 16,            Conv, 3, 1, 245, 735
 17,             Add, 2, 1, 735, 250
 18,            Relu, 1, 1, 250, 251
 19,            Conv, 3, 1, 251, 741
 20,            Relu, 1, 1, 741, 254
 21,            Conv, 3, 1, 254, 744
 22,             Add, 2, 1, 744, 257
 23,            Relu, 1, 1, 257, 258
 24,            Conv, 3, 1, 258, 753
 25,            Conv, 3, 1, 258, 747
 26,            Relu, 1, 1, 747, 261
 27,            Conv, 3, 1, 261, 750
 28,             Add, 2, 1, 750, 266
 29,            Relu, 1, 1, 266, 267
 30,            Conv, 3, 1, 267, 756
 31,            Relu, 1, 1, 756, 270
 32,            Conv, 3, 1, 270, 759
 33,             Add, 2, 1, 759, 273
 34,            Relu, 1, 1, 273, 274
 35,            Conv, 3, 1, 274, 406
 36,            Conv, 3, 1, 274, 768
 37,            Conv, 3, 1, 274, 762
 38,            Relu, 1, 1, 762, 277
 39,            Conv, 3, 1, 277, 765
 40,             Add, 2, 1, 765, 282
 41,            Relu, 1, 1, 282, 283
 42,            Conv, 3, 1, 283, 771
 43,            Relu, 1, 1, 771, 286
 44,            Conv, 3, 1, 286, 774
 45,             Add, 2, 1, 774, 289
 46,            Relu, 1, 1, 289, 290
 47,            Conv, 3, 1, 290, 777
 48,            Relu, 1, 1, 777, 293
 49,             Add, 2, 1, 293, 304
 50,            Conv, 3, 1, 304, 327
 51,         Reshape, 2, 1, 327, 334
 52,            Conv, 3, 1, 304, 314
 53,         Reshape, 2, 1, 314, 321
 54,            Conv, 3, 1, 304, 305
 55,         Reshape, 2, 1, 305, 312
 56,       Transpose, 1, 1, 312, 313
 57,          MatMul, 2, 1, 313, 322
 58,         Reshape, 2, 1, 322, 323
 59,         Softmax, 1, 1, 323, 324
 60,         Squeeze, 1, 1, 324, 325
 61,       Transpose, 1, 1, 325, 326
 62,          MatMul, 2, 1, 334, 335
 63,         Reshape, 2, 1, 335, 342
 64,             Mul, 2, 1, ortshared_6_1_1, 343
 65,             Add, 2, 1, 343, 344
 66,            Conv, 3, 1, 344, 780
 67,            Relu, 1, 1, 780, 347
 68,            Conv, 3, 1, 347, 783
 69,            Relu, 1, 1, 783, 350
 70,             Add, 2, 1, 350, 361
 71,            Conv, 3, 1, 361, 384
 72,         Reshape, 2, 1, 384, 391
 73,            Conv, 3, 1, 361, 371
 74,         Reshape, 2, 1, 371, 378
 75,            Conv, 3, 1, 361, 362
 76,         Reshape, 2, 1, 362, 369
 77,       Transpose, 1, 1, 369, 370
 78,          MatMul, 2, 1, 370, 379
 79,         Reshape, 2, 1, 379, 380
 80,         Softmax, 1, 1, 380, 381
 81,         Squeeze, 1, 1, 381, 382
 82,       Transpose, 1, 1, 382, 383
 83,          MatMul, 2, 1, 391, 392
 84,         Reshape, 2, 1, 392, 399
 85,             Mul, 2, 1, ortshared_8_1_1, 400
 86,             Add, 2, 1, 400, 401
 87,            Conv, 3, 1, 401, 786
 88,            Relu, 1, 1, 786, 404
 89,            Conv, 3, 1, 404, 407
 90,        Upsample, 2, 1, 407, 411
 91,             Add, 2, 1, 406, 412
 92,            Conv, 3, 1, 412, 419
 93,            Conv, 3, 1, 419, 423
 94,            Relu, 1, 1, 423, 424
 95,            Conv, 3, 1, 424, 425
 96,         Sigmoid, 1, 1, 425, 441
 97,             Add, 2, 1, 441, 443
 98,            Relu, 1, 1, 443, 444
 99,             Add, 2, 1, 444, 446
100,             Mul, 2, 1, 447, 448
101,             Add, 2, 1, 449, 450
102,            Relu, 1, 1, 450, 451
103,             Mul, 2, 1, 447, 453
104,             Add, 2, 1, 449, 455
105,         MaxPool, 1, 1, 455, 486
106,             Mul, 2, 1, 486, 488
107,             Add, 2, 1, 455, 489
108,             Mul, 2, 1, 489, 491
109,             Add, 2, 1, 491, 493
110,            Relu, 1, 1, 493, 494
111,             Mul, 2, 1, 494, 496
112,             Mul, 2, 1, 455, 497
113,         Squeeze, 1, 1, 497, 498
114,         Reshape, 2, 1, 498, 500

Input tensor name -  image 
Output tensor name - 258 
Output tensor name - 412 
Output tensor name - 419 
Output tensor name - 500 
*** In TIDL_createStateImportFunc *** 
Compute on node : TIDLExecutionProvider_TIDL_1_1
  0,            Conv, 3, 1, 419, 426
  1,            Relu, 1, 1, 426, 427
  2,            Conv, 3, 1, 427, 428
  3,         Reshape, 2, 1, 428, 473
  4,       Transpose, 1, 1, 473, 474
  5,         Reshape, 2, 1, 474, 485
  6,            Conv, 2, 1, 485, 511
  7,       Transpose, 1, 1, 511, 512
  8,         Reshape, 2, 1, 512, 514
  9,         Reshape, 2, 1, 502, seeds
 10,         Squeeze, 1, 1, seeds, 507
 11,          Gather, 2, 1, 514, 516
 12,       Transpose, 1, 1, 516, 528
 13,         Reshape, 2, 1, 528, 537
 14,            Conv, 2, 1, 537, 538
 15,       Transpose, 1, 1, 538, 539
 16,         Reshape, 2, 1, 539, 541
 17,            Conv, 3, 1, 258, 405
 18,        Upsample, 2, 1, 412, 416
 19,             Add, 2, 1, 405, 417
 20,            Conv, 3, 1, 417, 418
 21,            Conv, 3, 1, 418, 789
 22,            Relu, 1, 1, 789, 458
 23,            Conv, 3, 1, 458, 792
 24,            Relu, 1, 1, 792, 461
 25,            Conv, 3, 1, 461, 795
 26,            Relu, 1, 1, 795, 464
 27,          Concat, 2, 1, 523, 524
 28,         Reshape, 2, 1, 524, 527
 29,          MatMul, 2, 1, 541, 573
 30,         Reshape, 2, 1, 573, 585
 31,            Conv, 2, 1, 537, 542
 32,       Transpose, 1, 1, 542, 543
 33,         Reshape, 2, 1, 543, 545
 34,             Add, 2, 1, 545, 554
 35,         Reshape, 2, 1, 554, 570
 36,         Reshape, 2, 1, 570, 594

Input tensor name -  502 

Input tensor name -  419 

Input tensor name -  412 

Input tensor name -  258 
Output tensor name - seeds 
Output tensor name - 507 
Output tensor name - 485 
Output tensor name - 594 
Output tensor name - 527 
Output tensor name - 585 
*** In TIDL_createStateImportFunc *** 
Compute on node : TIDLExecutionProvider_TIDL_2_2
  0,            Conv, 2, 1, 485, 604
  1,       Transpose, 1, 1, 604, 605
  2,         Reshape, 2, 1, 605, 607
  3,          Gather, 2, 1, 607, 609
  4,       Transpose, 1, 1, 609, 621
  5,         Reshape, 2, 1, 621, 630
  6,            Conv, 2, 1, 630, 631
  7,       Transpose, 1, 1, 631, 632
  8,         Reshape, 2, 1, 632, 634
  9,          MatMul, 2, 1, 634, 666
 10,         Reshape, 2, 1, 666, 678
 11,            Conv, 2, 1, 630, 635
 12,       Transpose, 1, 1, 635, 636
 13,         Reshape, 2, 1, 636, 638
 14,             Add, 2, 1, 638, 647
 15,         Reshape, 2, 1, 647, 663
 16,         Reshape, 2, 1, 663, 687
 17,         Reshape, 2, 1, 595, masks
 18,       Transpose, 1, 1, masks, rout_tensor_1
 19,            Conv, 3, 1, rout_tensor_1, 710
 20,            Relu, 1, 1, 710, 711
 21,       Transpose, 1, 1, 711, 712
 22,            Conv, 3, 1, 712, 713
 23,       Transpose, 1, 1, 713, cls_logits
 24,            Conv, 3, 1, rout_tensor_1, v_c2_2_
 25,            Relu, 1, 1, v_c2_2_, rout_tensor_3
 26,            Conv, 3, 1, rout_tensor_3, v_c2_4_
 27,       Transpose, 1, 1, v_c2_4_, feat_range

Input tensor name -  595 

Input tensor name -  485 

Input tensor name -  507 

Input tensor name -  527 
Output tensor name - masks 
Output tensor name - feat_range 
Output tensor name - cls_logits 
Output tensor name - 687 
Output tensor name - 678 
*** In TIDL_createStateImportFunc *** 
Compute on node : TIDLExecutionProvider_TIDL_3_3
  0,         Reshape, 2, 1, 688, regs

Input tensor name -  688 
Output tensor name - regs 
 Graph Domain TO version : 9In TIDL_onnxRtImportInit subgraph_name=subgraph_0
Layer 0, subgraph id subgraph_0, name=258
Layer 1, subgraph id subgraph_0, name=412
Layer 2, subgraph id subgraph_0, name=419
Layer 3, subgraph id subgraph_0, name=500
Layer 4, subgraph id subgraph_0, name=image
In TIDL_runtimesOptimizeNet: LayerIndex = 120, dataIndex = 116 

 ************** Frame index 1 : Running float import ************* 
In TIDL_runtimesPostProcessNet 
In TIDL_runtimesPostProcessNet 1
In TIDL_runtimesPostProcessNet 2
In TIDL_runtimesPostProcessNet 3
INFORMATION: [TIDL_ResizeLayer] Upsample_143 Any resize ratio which is power of 2 and greater than 4 will be placed by combination of 4x4 resize layer and 2x2 resize layer. For example a 8x8 resize will be replaced by 4x4 resize followed by 2x2 resize.
****************************************************
**          1 WARNINGS          0 ERRORS          **
****************************************************
In TIDL_runtimesPostProcessNet 4
************ in TIDL_subgraphRtCreate ************ 
 The soft limit is 2048
The hard limit is 2048
MEM: Init ... !!!
MEM: Init ... Done !!!
 0.0s:  VX_ZONE_INIT:Enabled
 0.7s:  VX_ZONE_ERROR:Enabled
 0.8s:  VX_ZONE_WARNING:Enabled
 0.2973s:  VX_ZONE_INIT:[tivxInit:190] Initialization Done !!!
************ TIDL_subgraphRtCreate done ************ 
 
**********  Frame Index 1 : Running float inference **********
 Graph Domain TO version : 9In TIDL_onnxRtImportInit subgraph_name=subgraph_1
Layer 0, subgraph id subgraph_1, name=seeds
Layer 1, subgraph id subgraph_1, name=507
Layer 2, subgraph id subgraph_1, name=485
Layer 3, subgraph id subgraph_1, name=594
Layer 4, subgraph id subgraph_1, name=527
Layer 5, subgraph id subgraph_1, name=585
Layer 6, subgraph id subgraph_1, name=502
Layer 7, subgraph id subgraph_1, name=419
Layer 8, subgraph id subgraph_1, name=412
Layer 9, subgraph id subgraph_1, name=258
In TIDL_runtimesOptimizeNet: LayerIndex = 47, dataIndex = 41 
Warning : Requested Input Data Convert Layer is not Added to the network, It is currently not Optimal

 ************** Frame index 1 : Running float import ************* 
In TIDL_runtimesPostProcessNet 
In TIDL_runtimesPostProcessNet 1
In TIDL_runtimesPostProcessNet 2
In TIDL_runtimesPostProcessNet 3
INFORMATION: [TIDL_ResizeLayer] Upsample_145 Any resize ratio which is power of 2 and greater than 4 will be placed by combination of 4x4 resize layer and 2x2 resize layer. For example a 8x8 resize will be replaced by 4x4 resize followed by 2x2 resize.
****************************************************
**          1 WARNINGS          0 ERRORS          **
****************************************************
In TIDL_runtimesPostProcessNet 4
************ in TIDL_subgraphRtCreate ************ 
 ************ TIDL_subgraphRtCreate done ************ 
free(): double free detected in tcache 2
Aborted (core dumped)
log_int16.txt
0 Pratik Kedar over 1 year ago in reply to Deepanshu Yadvandu
TI__Mastermind 24041 points
Hi Deepanshu,
Is this the same model being discussed here : https://e2e.ti.com/support/processors-group/processors/f/processors-forum/1385387/tda4vm-edgeai-topk-custom-layer-implementation-review-required-from-quantization-perspective
Can you please help me understand the same.
Thanks
0 Deepanshu Yadvandu over 1 year ago in reply to Pratik Kedar
Prodigy 90 points
Yes its the same model
0 Pratik Kedar over 1 year ago in reply to Deepanshu Yadvandu
TI__Mastermind 24041 points
I would recommend to close the current thread as the same model is being discussed in above shared link, adding here for reference again.
Pratik Kedar said:
Is this the same model being discussed here : https://e2e.ti.com/support/processors-group/processors/f/processors-forum/1385387/tda4vm-edgeai-topk-custom-layer-implementation-review-required-from-quantization-perspective
An engineer is assigned for this lets continue discussion on the same thread, and avoid creating duplicates with almost no nuances.
Thank You
0 Deepanshu Yadvandu over 1 year ago in reply to Pratik Kedar
Prodigy 90 points
Hello Pratik,
The problems we have highlighted in both the tickets are different even though the model in use is the same.
I discussed the issue of multiple tickets with our managers. We cannot close the ticket without a solution. Please ask for the details you require and focus on solving the issue.
Also note that we are getting minimal response on the other tickets also.
Regards
0 Pratik Kedar over 1 year ago in reply to Deepanshu Yadvandu
TI__Mastermind 24041 points
Deepanshu,
I understand the situation and will make sure we help you in this case.
Looking at the logs provided above with tools version 9.2.9.0 seems like in one the four subgraphs during float inference the core dump issue occurred.
I will reverify the same observation at my end update the thread.
Thank You
0 Pratik Kedar over 1 year ago in reply to Pratik Kedar
TI__Mastermind 24041 points
Hi Deepanshu,
I didnt had chance to go through this thread as of now, am in quest of planning few threads to my ready queue for coming week.
Can you confirm this e2e is for which model ? lane det or other one ? We are currently supporting softmax e2e after our previous alignment.
For the given model have you modified any model import code ? Are there any delta wrt std tools offering here ?
Thanks
0 Deepanshu Yadvandu over 1 year ago in reply to Pratik Kedar
Prodigy 90 points
Hello Pratik,
This is regarding the lane detection model. Our current changes are not related to this issue.
0 Asha Bhandarkar over 1 year ago in reply to Deepanshu Yadvandu
TI__Genius 10170 points
From 9/5,
This is a priority for Lucid moving forward as it is on lane centering model.
- Asha
Processors

Processors forum

TDA4VM: Edgeai: Calibration issue with Batch Normalization layer implemented for constant multiply broadcast.
