Issues converting YOLOv8 ONNX model to uint8 .nb for STM32MP25

[20DeViL00]

 

Hi everyone,

I’m currently working on deploying a YOLOv8 model on the STM32MP25 using ST Edge AI. Here’s what I’ve done so far and where I’m stuck.

Training & Export:

from ultralytics import YOLO

model = YOLO("yolov8n.pt") # medium variant

 

model.train( data=path, epochs=50, imgsz=256, project="Training", name="yolov8n_run", save=True )

 

# Export ONNX

model.export( format="onnx", imgsz=256, opset=10, dynamic=False, simplify=True, half=False, int8=False )

ONNX model details:

• Input: [1, 3, 256, 256], float32

• Output: [1, 5, 1344], float32

I successfully converted it to .nb using:

stedgeai generate --model best.onnx --target stm32mp25

This produces a model in float16 (I tried to run this but it did not work).

Problem:
STM32MP2 works with uint8 models, so I tried converting input/output to uint8:

 

 

stedgeai generate --model best.onnx --target stm32mp25 --input-data-type uint8 --output-data-type uint8

This fails during compilation with:

Fatal model compilation error: 512 Error during NBG compilation, model is not supported

Observation:

• The normally .nb file is generating, without conversion.

• Explicitly trying uint8 fails.

• The error suggests the ST Edge AI compiler doesn’t support direct uint8 conversion for this ONNX model.

• I've also tried ove the ST Edge AI developer cloud, but did not worked.

My Question:
Has anyone successfully converted a YOLOv8 ONNX model to uint8 for STM32MP25? Are there recommended steps for preparing the model for uint8 quantization?

Any advice on preprocessing, exporting, or using ST tools to get a uint8-ready model would be highly appreciated.

Thanks!
Aman

[Julian E.]

Hello @20DeViL00,

 

You need to use an int8 model.

In your command:

model.export( format="onnx", imgsz=256, opset=10, dynamic=False, simplify=True, half=False, int8=True )

 

The AI MP2 team also recommend to use Tflite if you can, as it seems to work better in most case:

model.export( format="tflite", ... , int8=True )

 

Then you can do your generate.

 

The --input-data-type uint8 --output-data-type uint8 only add layers to fit the type you want in input and output. It does not change the model.

 

Have a good day,

Julian

[20DeViL00]

 

Hi @Julian E.,

Thanks for the clarification regarding the int8 export. I tried following your suggestion. Here’s what I did:

model.train(data=path, epochs=100, imgsz=256, project="Training", name="yolov8n_run", save=True, int8=True)

model.export( format="onnx", imgsz=256, opset=10, dynamic=False, simplify=True, half=False, int8=True )

Then I ran the ST Edge AI generate command:

stedgeai generate --model best.onnx --target stm32mp25 --input-data-type uint8 --output-data-type uint8

But I’m hitting this error:

ST Edge AI Core v2.2.0-20266 2adc00962 make: *** [.../export_ovxlib/makefile.linux:53: vnn_pre_process.o] Error 1 E 17:34:11 Fatal model compilation error: 512 E 17:34:11 ('Fatal model compilation error: 512', 'nbg_compile') Error during first compilation. Retrying with other settings... make: *** [.../export_ovxlib/makefile.linux:53: vnn_pre_process.o] Error 1 E 17:34:13 Fatal model compilation error: 512 E 17:34:13 ('Fatal model compilation error: 512', 'nbg_compile') E010(InvalidModelError): Error during NBG compilation, model is not supported

It seems the model isn’t compiling for STM32MP2. 

Thanks again for the guidance!

Aman Sharma

[Julian E.]

Hello @20DeViL00,

 

It seems that in ONNX the arg int8 do not work ( https://docs.ultralytics.com/modes/export/#export-formats).

 

So you can try to use TFLITE instead or get your non quantized model, then quantize it manually before doing the generate.

 

Have a good day,

Julian