Build failed error when deploying pose detection model

[athern27]

Hi everyone, 
I am trying to deploy a pose detection model on stm32n6570-dk board but I am getting the following error.

(st_zoo) kartikkhandewal@ATL-HPZG14-99:~/stm32packages/stm32ai-modelzoo-services/pose_estimation/src$ python3 stm32ai_main.py --config-path ./config_file_examples --config-name deployment_n6_yolo_mpe_config.yaml
WARNING: All log messages before absl::InitializeLog() is called are written to STDERR
E0000 00:00:1740729082.336274  136562 cuda_dnn.cc:8310] Unable to register cuDNN factory: Attempting to register factory for plugin cuDNN when one has already been registered
E0000 00:00:1740729082.339482  136562 cuda_blas.cc:1418] Unable to register cuBLAS factory: Attempting to register factory for plugin cuBLAS when one has already been registered
[WARNING] The usable GPU memory is unlimited.
Please consider setting the 'gpu_memory_limit' attribute in the 'general' section of your configuration file.
[INFO] : Running `deployment` operation mode
[INFO] : The random seed for this simulation is 123
INFO: Created TensorFlow Lite XNNPACK delegate for CPU.
[INFO] : Generating C header file for Getting Started...
loading model.. model_path="yolov8n_256_quant_pc_uf_pose_pallet.tflite"
loading conf file.. "../../application_code/pose_estimation/STM32N6/stmaic_STM32N6570-DK.conf" config="None"
"n6 release" configuration is used
[INFO] : Selected board :  "STM32N6570-DK Getting Started Pose Estimation (STM32CubeIDE)" (stm32_cube_ide/n6 release/stm32n6)
[INFO] : Compiling the model and generating optimized C code + Lib/Inc files:  yolov8n_256_quant_pc_uf_pose_pallet.tflite
setting STM.AI tools.. root_dir="", req_version=""
 Cube AI Path: "/home/kartikkhandewal/STEdgeAI/2.0/Utilities/linux/stedgeai".
[INFO] : Offline CubeAI used; Selected tools:  10.0.0 (x-cube-ai pack)
loading conf file.. "../../application_code/pose_estimation/STM32N6/stmaic_STM32N6570-DK.conf" config="None"
"n6 release" configuration is used
compiling... "yolov8n_256_quant_pc_uf_pose_pallet_tflite" session
 model_path  : ['yolov8n_256_quant_pc_uf_pose_pallet.tflite']
 tools       : 10.0.0 (x-cube-ai pack)
 target      : "STM32N6570-DK Getting Started Pose Estimation (STM32CubeIDE)" (stm32_cube_ide/n6 release/stm32n6)
 options     : --st-neural-art default@../../application_code/pose_estimation/STM32N6/Model/user_neuralart.json --input-data-type uint8 --inputs-ch-position chlast
"series" value is not coherent.. stm32n6 != stm32n6npu
 results -> RAM=1,675,264 IO=196,608:220,416 WEIGHTS=3,243,377 MACC=0 RT_RAM=1,893 RT_FLASH=504,138 LATENCY=0.000
[INFO] : Optimized C code + Lib/Inc files generation done.
[INFO] : Building the STM32 c-project..
deploying the c-project.. "STM32N6570-DK Getting Started Pose Estimation (STM32CubeIDE)" (stm32_cube_ide/n6 release/stm32n6)
updating.. n6 release
 -> s:copying file.. "network.c" to ../../application_code/pose_estimation/STM32N6/Model/network.c
 -> s:copying file.. "network_ecblobs.h" to ../../application_code/pose_estimation/STM32N6/Model/network_ecblobs.h
 -> s:copying file.. "network_atonbuf.xSPI2.raw" to ../../application_code/pose_estimation/STM32N6/Model/network_atonbuf.xSPI2.raw
 -> s:removing dir.. ../../application_code/pose_estimation/STM32N6/Middlewares/AI_Runtime/Lib/GCC/ARMCortexM55
 -> s:copying dir.. "ARMCortexM55" to ../../application_code/pose_estimation/STM32N6/Middlewares/AI_Runtime/Lib/GCC/ARMCortexM55
 -> s:removing dir.. ../../application_code/pose_estimation/STM32N6/Middlewares/AI_Runtime/Inc
 -> s:copying dir.. "Inc" to ../../application_code/pose_estimation/STM32N6/Middlewares/AI_Runtime/Inc
 -> s:removing dir.. ../../application_code/pose_estimation/STM32N6/Middlewares/AI_Runtime/Npu/ll_aton
 -> s:copying dir.. "ll_aton" to ../../application_code/pose_estimation/STM32N6/Middlewares/AI_Runtime/Npu/ll_aton
 -> u:copying file.. "app_config.h" to ../../application_code/pose_estimation/STM32N6/Inc/app_config.h
 -> updating cproject file "/home/kartikkhandewal/stm32packages/stm32ai-modelzoo-services/application_code/pose_estimation/STM32N6/STM32CubeIDE" with "NetworkRuntime1000_CM55_GCC.a"
building.. n6 release
[returned code = 1 - FAILED]
flashing.. n6 release STM32N6570-DK
[returned code = 1 - FAILED]
Board programming failed: " Error: binary file does not exist:  Debug/STM32N6_GettingStarted_PoseEstimation.bin"
[returned code = 1 - FAILED]
Board programming failed: "Error: File does not exist: STM32N6_GettingStarted_PoseEstimation_signed.bin"
Board programming failed: "Error: File does not exist: STM32N6_GettingStarted_PoseEstimation_signed.bin"
Board programming failed: "Error: File does not exist: STM32N6_GettingStarted_PoseEstimation_signed.bin"
[INFO] : Deployment complete.
[INFO] : Please on STM32N6570-DK toggle the boot switches to the left and power cycle the board.

I have deployed object detection models on board and they have worked fine but I am not able to deploy this.
Steps I followed-
best.pt ----> my original pose detection model (12 keypoints to detect)
converted it to tflite which gave me saved_model using 

from ultralytics import YOLO

model_path = 'best.pt'
model = YOLO(model_path)
results = model.export(format='tflite', int8=True, imgsz=[256, 256])

Quantized model following this tutorial .

Deployed it using this tutorial 

Kindly help.

[Julian E.]

Hello @athern27 ,

The issue may come from the 12 keypoints to detect. It seems that we only support 13 and 17.

Can you try to use a model with 13 or 17 keypoints? 

You can find ultralytics yolos model examples here:

ultralytics/examples/YOLOv8-STEdgeAI at main · stm32-hotspot/ultralytics

If it doesn't help, it may be an issue in the deployment scripts, so do not hesitate to come back to me if you still have an issue.

Have a good day,

Julian

[athern27]

Hi @Julian E. , 

I tried running the yolov8n-pose.pt model, which is designed to detect 17 body keypoints. After quantizing it, I was able to run it successfully with good accuracy.

Is there no way to deploy a 12-keypoint model on the board?

Kindly help.

[Julian E.]

Hello @athern27,

I am not exactly sure about the procedure, could you please try to adapt the postprocess_conf.h with​

• #define AI_MPE_YOLOV8_PP_TOTAL_BOXES (1200)​
• #define AI_POSE_PP_POSE_KEYPOINTS_NB (12)​

And look at the need app.c and see if you see something related to that.

I've got the information that it is not a big deal to change the getting started but I don't have clear guideline.

If you can generate a model with 17 and 13 keypoints and compare what is different, you may be able to understand the part related to the keypoints.

Have a good day,

Julian