Custom Network Quantization Method for Lightweight CNN Acceleration on FPGAs

Table 9

The ablation experiment 2.

Network	Number of preceding convolution op	Top-1 accuracy (%)	FPS
	0	94.49	18.50
	9	94.36	17.06
	16	94.21	31.51
	20	94.10	36.29
	1	94.49	19.28
	46	93.84	52.94