Multichannel Cross-Scale Semantic Coherent Attention Network for Image Inpainting

Table 4

Ablation experiment result.

Method	Res	Attention	PSNR↑	SSIM↑	L1↓
RES1ATT1	√	√	37.6460	0.9794	0.0100
	Improvement compared with baseline		10.50%	0.13%	47.09%
RES0ATT1	×	√	35.8867	0.9767	0.0125
	Improvement compared with baseline		5.34%	−0.14%	33.86%
RES1ATT0	√	×	35.2383	0.9771	0.0137
	Improvement compared with baseline		3.43%	−0.10%	27.51%
RES0ATT0	×	×	34.0685	0.9781	0.0189
			—	—	—

↑ means the bigger, the better; ↓ means the smaller, the better.