An FPGA-Based Hardware Accelerator for CNNs Using On-Chip Memories Only: Design and Benchmarking with Intel Movidius Neural Compute Stick

Figure 5

Accuracy and MSE to the change of the number of bits of filter elements.