Kunal’s Status Report (12/4)

We finished up benchmarking the latencies of our fsrcnn model and I wrote a profiler that would take upscaled frames and build metrics based on the quality of these images. This we will use in the final demo to compare and contrast latencies of the software model running on the hardware device. The nature of the computation involves a scan over the pixel set and a buffering mechanism for the previous pixels fed into this pipeline.  We are actively tuning the model and analyzing our results, we expect to get to a significantly lower latency and be able to profile the results much more efficiently.