er1cz / deploying_cnn_on_fpga_using_opencl Goto Github PK

hello everyone, when deploying the object recognition application on the intel Arria 10 soc card,
I had to create the bitstream file for my card with the offline compiler, and then you did not include the make I use a make that I found in the intels examples.
when running the application on the map I get warnings about data alignment transfer from the host part to the FPGA part.

I saw in some forum that I can use the function posix_memalign, and looking in the project folders, I found the data alignment function of the host part:
`//////////////////////////////////////////
// Host allocation functions for alignment
//////////////////////////////////////////
// This is the minimum alignment requirement to ensure DMA can be used.
const unsigned AOCL_ALIGNMENT = 64;

#ifdef _WIN32 // Windows
void *alignedMalloc(size_t size) {
return _aligned_malloc (size, AOCL_ALIGNMENT);
}
void alignedFree(void * ptr) {
_aligned_free(ptr);
}
#else // Linux
void *alignedMalloc(size_t size) {
void *result = NULL;
int rc;
rc = posix_memalign (&result, AOCL_ALIGNMENT, size);
return rc;
}
void alignedFree(void * ptr) {
free (ptr);
}
#endif
`
but this function is not used in the main()
so, how can I remove these warning?
thanks for any help!

I got error while trying to execute the above command on the DE10 Nano board.

Hey guys.
I have the de1soc board and i like compile this CNN in my board.
For this, i need generate the aocx with my makefile.
How can I do this?

Hi

I tried to run your example on an Intel CPU. Using the version 1.3 OpenCL kernel, the prediction result is

predicted label: n01440764 tench, Tinca tinca

Could you please describe the process to convert an jpg image to a header file (e.g. dog.h) included in your main.cpp ?

Thank you for your attention

Hello , great work Er1CZ , thanks for this git repo, i get the following error can you please look in the error and guide me.

Hi, I am using your codes as reference to do my version of CNN, but I faced a problem when compiling the host program. In your host program, there are variables such as sample, conv1_weight used, but I noticed that these are declared in another 2 files, which are dog.h and squeezenet_params.h. Can anyone tell me what should I prepare if I need to do my own CNN for object detection? I want to do the similar thing for SqueezeDet, what are the steps I need to go through?

Hey guys!
What i can test 1 image ?
@Er1cZ