Part Number: TDA4VM
I am facing an issue with TIDL import tool.
I have created a tftite model by training my own data set.
I m trying to import that model into TIDL framework and facing issues in importing.
The model created will be used for wakeword detection and it is trained using audio data.
Here are the params I m using to import the tflite model.
modelType = 3
numParamBits = 15
quantizationStyle = 2
inputNetFile = /home/jithin/project/ava/wakeword_training/tidl_convert/wakeword_int8.tflite
outputNetFile = "/home/jithin/project/ava/wakeword_training/tidl_convert/wakeword_.bin"
outputParamsFile = "/home/jithin/project/ava/wakeword_training/tidl_convert/wakeword__"
quantizationStyle = 0
numParamBits = 8
numFeatureBits = 8
# Input tensor information (NOT image!)
inWidth = 13
inHeight = 101
inNumChannels = 1
inElementType = 1
NetInElementType = 1
inDataNorm = 0
inData = dummy_input.bin
postProcType = 0
When I run the command as "./tidl_model_import.out tidl_import_config.txt"
Below logs are coming,
========================= [Model Compilation Started] =========================
Model compilation will perform the following stages:
1. Parsing
2. Graph Optimization
3. Quantization & Calibration
4. Memory Planning
============================== [Version Summary] ==============================
-------------------------------------------------------------------------------
| TIDL Tools Version | 11_02_04_00 |
-------------------------------------------------------------------------------
| C7x Firmware Version | 11_02_04_00 |
-------------------------------------------------------------------------------
TFLite model (Flatbuf) file : /home/jithin/project/ava/wakeword_training/tidl_convert/wakeword_int8.tflite
TIDL network file : /home/jithin/project/ava/wakeword_training/tidl_convert/wakeword_.bin
TIDL IO info file : /home/jithin/project/ava/wakeword_training/tidl_convert/wakeword__
Number of operators: 11
Segmentation fault (core dumped)
It resulted in a segmentation fault.
When I run using gdb, here is what I found;
Program received signal SIGSEGV, Segmentation fault.
0x00007fffe9425176 in TIDL_tfliteCopyInputConstTensor(tflite::Model const*, int, int, sBuffer_t&, sBuffer_t&, sBuffer_t&) [clone .cold] ()
Is it something wrong with my config file or am I missing something in TIDL model import ?
