Loading a ultralytics yolov8 model on luckfox pico ultra w

2025-07-14 6:04

Hi,
I am trying to convert a ultralytics yolov8 model to run on my luckfox pico ultra w but it is giving this issue
conversion i have done on my ubuntu system

my approach => yolov8n.pt ----converted to----> yolov8.onnx ----converted to----> yolov8.rknn

later this .rknn model i am using but facing this issue

E RKNN: failed to submit!, op id: 133, op name: exSoftmax13:/model.22/dfl/Softmax, task start: 693, task number: 67538, run task counter: 1972, int status: 0
rknn_run fail! ret=-1
init_yolov8_model fail! ret=-1

my questions
1] Do we need to tweak or make any changes in the model before using it with luckfox pico?
2] Can you please provide the .pt file used in these examples on the official site https://wiki.luckfox.com/Luckfox-Pico/L ... RKNN-Test/

Thanks in advance for the help

PFA the screenshot

2025-07-14 13:29

Hello, before running RKNN, the structure of the model and the required operators need to be determined. If the source code for training the model and the source code for inference do not match, the model may not be able to run (for example, both are YOLOv8 models but different post-processing methods in the source codes will cause the model to be unusable).

The problem you encountered is likely due to incompatible operators. Since there are many types of convolutional neural networks with complex structures, we do not have professional engineers to handle this, so we do not provide technical support for the use of custom models. It is recommended to consult the RKNN official.

The .pt file mentioned in the wiki, we have provided the relevant git repository address. You can obtain it from the specified repository.

2025-07-14 13:49

Thanks for the response
Can you please share the link for the git repository to get the .pt file because I was not able to find it.

It would of great help if you kindly the share the git link for obtaining the .pt files

Thanks in advance