Rknn model zoo fails during conversion with core dump message

2024-08-15 13:50

Greetings.
I am following the instructions to get a custom yolov5 converted to rknn for use with a Luckfox Pico max.
All the versions are according to spec and when i try to convert any yolov5, including the default coco one, i am getting illegal isntrction/core dump errors.
See messages bellow

(rockchip-rknn) stelios@Dev-VirtualBox:~/projects/luckfox/rknn_model_zoo/examples/yolov5/python$ python convert.py yolov5s.onnx rv1106
W __init__: rknn-toolkit2 version: 1.6.0+81f21f4d
--> Config model
done
--> Loading model
W load_onnx: It is recommended onnx opset 19, but your onnx model opset is 12!
W load_onnx: Model converted from pytorch, 'opset_version' should be set 19 in torch.onnx.export for successful convert!
Loading : 100%|█████████████████████████████████████████████████| 135/135 [00:00<00:00, 5606.30it/s]
done
--> Building model
W build: found outlier value, this may affect quantization accuracy
const name abs_mean abs_std outlier value
onnx::Conv_371 0.83 1.39 14.282
Illegal instruction (core dumped)

Any idea what the problem is ?

2024-08-16 2:07

Hello, it seems that the input parameters for your command are incorrect and lack the quantization option. Here is the command I executed in the same directory based on the official documentation for your reference.

Code: Select all

python convert.py ../model/yolov5s_relu.onnx rv1106 i8 ../model/yolov5s_relu.rknn

2024-08-16 7:20

Thanks for the response.
The problem was that the ubuntu VM i was using was missing the AVX cpu command and most/all of AI software these days requires it (tensorflow , pytorch etc)
Enabling the command resolved it.
I think there should be a mention of it on the wiki, as during my searches i found a lot of people with the same problem

2024-08-16 8:13

Thank you for your feedback. We will update the WIKI after verification.