[Retracted] A Visual Comprehensive Evaluation of Cultural and Creative Industry Projects Based on Improved Literature Measurement Method

Table 4

Layer by layer compression results of vgg-16 network with compression rate of 50%.

Convolution layer	Number of convolution kernels (before/after deletion)		Flip (before/after deletion)		Top1 precision (after deletion)
					Weights Sum|	Car	Ours	Fine tuning
Block1Conv1	64	32	4.8 × 10^6	1.3 × 10^6	0.7288	0.8	0.824	0.8718
Block1Conv2	64	32	4.8 × 10^6	1.3 × 10^6	0.637	0.3426	0.7471	0.8683
Block2Conv1	128	64	4.5 × 10^6	1.2 × 10^6	0.7175	0.6444	0.8014	0.8679
Block2Conv2	128	64	4.5 × 10^6	1.2 × 10^6	0.4676	0.5985	0.7764	0.867
Block3Conv1	256	128	4.3 × 10^6	1.1 × 10^6	0.7532	0.5134	0.8012	0.8665
Block3Conv2	256	128	4.3 × 10^6	1.1 × 10^6	0.6967	0.6574	0.7929	0.862
Block3Conv3	256	128	4.3 × 10^6	1.1 × 10^6	0.6926	0.5289	0.7529	0.8526
Block4Conv1	512	256	4.3 × 10^6	1.1 × 10^6	0.7483	0.5807	0.7709	0.8513
Block4Conv2	512	256	4.3 × 10^6	1.1 × 10^6	0.669	0.5604	0.7332	0.8498
Block4Conv3	512	256	4.3 × 10^6	1.1 × 10^6	0.5561	0.4674	0.775	0.8384
Block5Conv1	512	256	1.1 × 10^6	2.7 × 10^5	0.8086	0.5947	0.8253	0.8386
Block5Conv2	512	256	1.1 × 10^6	2.7 × 10^5	0.7695	0.5967	0.8208	0.8375
Block5Conv3	512	256	1.1 × 10^6	2.7 × 10^5	0.7583	0.5867	0.8262	0.8818