RCANResidualGroup代码实现

 RCAB 模块参考 https://blog.csdn.net/qq_41251963/article/details/120195167

## Residual Group (RG)
class ResidualGroup(nn.Module):def __init__(self, conv, n_feat, kernel_size, reduction, act, res_scale, n_resblocks):super(ResidualGroup, self).__init__()modules_body &＃61; []modules_body &＃61; [RCAB(conv, n_feat, kernel_size, reduction, bias&＃61;True, bn&＃61;False, act&＃61;nn.ReLU(True), res_scale&＃61;1) \for _ in range(n_resblocks)]modules_body.append(conv(n_feat, n_feat, kernel_size))self.body &＃61; nn.Sequential(*modules_body)def forward(self, x):res &＃61; self.body(x)res &＃43;&＃61; xreturn res


## Residual Channel Attention Network (RCAN)
class RCAN(nn.Module):def __init__(self, args, conv&＃61;common.default_conv):super(RCAN, self).__init__()n_resgroups &＃61; args.n_resgroupsn_resblocks &＃61; args.n_resblocksn_feats &＃61; args.n_featskernel_size &＃61; 3reduction &＃61; args.reduction scale &＃61; args.scale[0]act &＃61; nn.ReLU(True)# RGB mean for DIV2Krgb_mean &＃61; (0.4488, 0.4371, 0.4040)rgb_std &＃61; (1.0, 1.0, 1.0)self.sub_mean &＃61; common.MeanShift(args.rgb_range, rgb_mean, rgb_std)# define head modulemodules_head &＃61; [conv(args.n_colors, n_feats, kernel_size)]# define body modulemodules_body &＃61; [ResidualGroup(conv, n_feats, kernel_size, reduction, act&＃61;act, res_scale&＃61;args.res_scale, n_resblocks&＃61;n_resblocks) \for _ in range(n_resgroups)]modules_body.append(conv(n_feats, n_feats, kernel_size))# define tail modulemodules_tail &＃61; [common.Upsampler(conv, scale, n_feats, act&＃61;False),conv(n_feats, args.n_colors, kernel_size)]self.add_mean &＃61; common.MeanShift(args.rgb_range, rgb_mean, rgb_std, 1)self.head &＃61; nn.Sequential(*modules_head)self.body &＃61; nn.Sequential(*modules_body)self.tail &＃61; nn.Sequential(*modules_tail)def forward(self, x):x &＃61; self.sub_mean(x)x &＃61; self.head(x)res &＃61; self.body(x)res &＃43;&＃61; xx &＃61; self.tail(res)x &＃61; self.add_mean(x)return x