SK-TDA4VM: I use edgeai-tidl-tools and edgeai-benchmark successfully to to compile the model， but I fail to execute the model using app_edgeai.py.

He Gangqiang

Part Number: SK-TDA4VM

I have some problems：

（1）when I use app_edgeai.py to execute models, The following problems occur （models from modelzoo）

（2）I added "mean:" and “scale:” in "preprocess" of yaml file，Does it have a big impact on model execution?

preprocess:
  add_flip_image: false
  backend: pil
  crop: 224
  data_layout: NHWC
  interpolation: null
  pad_color: 0
  resize: 256
  resize_with_pad: false
  reverse_channels: false
  mean:
  - 128.0
  - 128.0
  - 128.0
  scale:
  - 0.007812
  - 0.007812
  - 0.007812

（3）Only one of the multiple models I've tested so far has been successful. it is “mnasnet_1.0_224.tflite”

over 3 years ago

+1 Takuma Fujiwara over 3 years ago

TI__Guru 50403 points

Hello,

There was a new release for the 8.2 version of the SDK yesterday. If you don't mind, could you try running the demos with the latest release to see if this error persists?

https://www.ti.com/tool/download/PROCESSOR-SDK-LINUX-SK-TDA4VM

Regards,

Takuma

0 He Gangqiang over 3 years ago in reply to Takuma Fujiwara

Prodigy 145 points

Hello，

the 8.2 version of the SDK solved my problem. Thanks again.

Regards,

gangqiang

0 He Gangqiang over 3 years ago in reply to Takuma Fujiwara

Prodigy 145 points

Hello,

I use "run_custom_pc.sh" to compile models (ONR-CL, ONR-OD, TFL-CL, TFL-OD) and I use app_edgeai.py to execute these models. I have the following situation, all the CL models tested can be executed successfully, but most of the OD models can't be executed.
So I want to know,
(1) Does the OD model need to set some relevant parameters? How to set it up?
(2) How to set "pipeline_configs" correctly? Is there a reference example?
(3) Where can I view the log of app_edgeai.py program execution?

Regards,

gangqiang

+1 Taekmin Lee over 3 years ago in reply to He Gangqiang

Expert 1445 points

Hi gangqiang,

1. In my experience, OD models compiled with TIDL_TOOLS do not work on EdgeAI Apps(both C++ and Python App).
To use OD models in EdgeAI Apps, try model compile using EdgeAI Benchmark.
Check this docs: https://github.com/TexasInstruments/edgeai-benchmark/blob/master/docs/custom_models.md

2. An example of "pipeline_config" in EdgeAI Benchmark can be found here.
https://github.com/TexasInstruments/edgeai-benchmark/blob/master/configs/detection.py

3. You can use below options on python based EdgeAI Apps.
-v : Verbose option to print profile info on stdout (default: Disabled)
-n : Disable curses report (default: Disabled)

Regards,
Lee

Processors

Processors forum

SK-TDA4VM: I use edgeai-tidl-tools and edgeai-benchmark successfully to to compile the model， but I fail to execute the model using app_edgeai.py.