Glad you enjoyed the read!
I am actually working with FPGA acceleration more and more lately, the DeePHI ip core looks really really nice, but for some reason, at this moment I am still stuck running it in the Zynq-7000, although it states it should also run there. And kind of unfortunate that it…
Glad you enjoyed it.
I am doing tests with tensorRT, which is obviously faster, and I am planning to update this with these numbers. But is also a lot more complex to get running. The comparison was made with what I could get running in a few seconds based on examples.