KL630 Custom Model Pre-processing
I have problem on creating custom model, I follow every tutorial (https://doc.kneron.com/docs/#plus_c/feature_guide/customized_api/create_kl630_single_model_example/)
but when running the solution_host_stream firmware on the device got this error:
I used yolov5 with custom class, I think the problem is not in NEF or hardware supported operators related stuff model, refer to this attachment
I try to search for custom preprocessing implementation but cannot find anywhere in the docs, could you help me on this?
The discussion has been closed due to inactivity. To continue with the topic, please feel free to post a new discussion.
Comments
Hi Alfi,
Based on the error log you provided, 'exist cpu op !!!,' it indicates that there are unsupported operators for the KL630 in the model.
The recommended process of model deployment is as follows. You can first verify the E2E simulation of NEF to confirm your model.
1. Verify the E2E simulation of NEF on the toolchain docker.
2. Confirm the results of KL630 companion mode.(python code)
3. Confirm the results of KL630 companion mode.(C code)
4. Move C code to KL630.
Ref. https://doc.kneron.com/docs/#toolchain/manual_1_overview/#_top
You're right thank you for the guide, I tried to change some model operators and now it works on stream but I have another issue here:
The model is running on solution_host_stream but the prediction results show random values, I tried to check from my customized post-processing function which registered in the VMF_NNM_INFERENCE_APP_CONFIG_T by adding print/logging function but the values are not being printed in the terminal....
I assume the post-processing function has never been called, or there is a specific way to call the post-processing function?
thank you in advance
Hi Alfi,
Please follow the previously mentioned model development steps for verification.
The first step will help ensure that your model is correct during the conversion process.
1. Verify the E2E simulation of NEF on the toolchain docker.
2. Confirm the results of KL630 companion mode.(python code)
3. Confirm the results of KL630 companion mode.(C code)
4. Move C code to KL630.
Thanks!