We are going to run our algorithm on A10 with OpenVino tool. I studied the tool and can run the sample on our A10 board.
My question is that if we change the deep learning model such as we use two models, shall we use a different .aocx file from that in the release package? If the answer is yes, is there a reference to change the file?
Thanks for your reply.
We are still studying the tool and need more time to do the try. I think my question is a common issue. If the developer change the model or develop a different model based on TensorFlow or others frameworks, how to download the .aocx file with the openVino tool? Will the developer write the opencl code and generate the .aocx?
Unfortunately, there isn't a way to edit the pre-compiled bit streams(.aocx) that come with OpenVINO underneath the Deployment DLA license that comes in the package. However, there is an opportunity to get a Development license(which comes with a fee) to edit the bitstreams and recompile them with specific architecture that fits your models to get better performance. Let me know if you are interested in the Development License?
Perhaps there is a little bit of misunderstanding. You don't strictly need to change aocx file if you change model.
This should only happen if you are missing some layers in HW or can use different aocx to get better performance.
No strict need to write own aocx.
Now， I am testing opemvino with our A10 board，the *.aocx which openvino tool provide do not match our board，so I want compile *.aocx by myself, but I can not find some doc for this, can you give me some suggestions？