SDAA177 November   2025 AM62A7 , AM67A , TDA4AL-Q1 , TDA4VE-Q1 , TDA4VEN-Q1 , TDA4VH-Q1 , TDA4VM , TDA4VP-Q1

 

  1.   1
  2.   Abstract
  3.   Trademarks
  4. 1Introduction
  5. 2C7xMMA DDR Usage Analysis and Optimization
  6. 3Testing Based on J722S and MobileNet
    1. 3.1 Firmware Compilation and Environment Configuration
    2. 3.2 Model Export and On-Board Inference
    3. 3.3 Memory Statistics
    4. 3.4 Modifying the Memory Map
    5. 3.5 Recompiling SDK and Updating to Board
    6. 3.6 On-Board Testing
  7. 4Summary
  8. 5References

Model Export and On-Board Inference

After completing 3.1, the default test model MobileNet will be installed in the SDK. Run in ti-processor-sdk-rtos-j722s-evm-11_01_00_04/c7x-mma-tidl/ti_dl/utils/tidlModelImport:

./out/tidl_model_import.out ../../test/testvecs/config/import/public/caffe/tidl_import_mobilenet_v1.txt

After the import completes, the model artifacts must be copied to the target board. Copy the exported model to the board, configure the inference parameters in infer.txt. The infer.txt file should contain the following configuration:

inFileFormat = 1
numFrames = 1
postProcType = 1
postProcDataId = 0
inData = input.bin
outData = "output.bin"
netBinFile = tidl_net_mobilenet_v1.bin
ioConfigFile = tidl_io_mobilenet_v1_1.bin
writeTraceLevel = 0
debugTraceLevel = 2
coreNum = 2

Please note to set debugTraceLevel = 2. The paths in this infer.txt file assume the model artifacts (tidl_net_mobilenet_v1.bin and tidl_io_mobilenet_v1_1.bin), and input.bin are in the same directory as the TI_DEVICE_armv8_test_dl_algo_host_rt.out executable.

Then run:

/opt/tidl_test/TI_DEVICE_armv8_test_dl_algo_host_rt.out s:infer.txt

Run /opt/vision_apps/vision_apps_init.sh to get the result in Figure 1.