{"id":6419,"date":"2022-08-28T16:58:36","date_gmt":"2022-08-28T08:58:36","guid":{"rendered":"http:\/\/139.9.1.231\/?p=6419"},"modified":"2022-08-28T16:58:37","modified_gmt":"2022-08-28T08:58:37","slug":"sunet-swin-transformer","status":"publish","type":"post","link":"http:\/\/139.9.1.231\/index.php\/2022\/08\/28\/sunet-swin-transformer\/","title":{"rendered":"SUNet: Swin Transformer with UNet for Image Denoising"},"content":{"rendered":"\n

ISCAS 2022\u7684\u4e00\u7bc7\u6587\u7ae0\uff0c\u4f5c\u4e3a\u9996\u4e2aSwin Transformer\u5728\u56fe\u50cf\u53bb\u566a\u9886\u57df\u7684\u5e94\u7528\uff0c\u6548\u679c\u6765\u8bf4\u611f\u89c9\u8fd8\u6709\u5f88\u5927\u63d0\u9ad8\u7a7a\u95f4\u3002\u4f46\u4e0d\u7684\u4e0d\u8bf4\uff0c\u81ea\u4eceSwin Transformer\uff082021\uff09\u63d0\u51fa\u540e\uff0c\u5728\u6574\u4e2acv\u9886\u57df\u72ec\u9886\u98ce\u9a9a\u3002\u4f5c\u4e3a\u4e00\u4e2a\u901a\u7528\u7684\u67b6\u6784\uff0c\u53ef\u4ee5\u5c06\u5176\u5e94\u7528\u5728\u5404\u4e2acv\u9886\u57df\uff0c\u4ecepaperwithcode\u91cc\u5c31\u53ef\u4ee5\u89c1\u5176\u5f71\u54cd\u529b\uff1a\uff08\u622a\u6b62\u523022.8.28\uff09<\/p>\n\n\n\n

1\u3001\u76ee\u6807\u68c0\u6d4b\uff1a<\/p>\n\n\n\n

\"\"<\/figure>\n\n\n\n

2\u3001\u56fe\u50cf\u8d85\u5206\u8fa8\u7387<\/p>\n\n\n\n

\"\"<\/figure>\n\n\n\n

3\u3001\u5b9e\u4f8b\u5206\u5272\uff1a<\/p>\n\n\n\n

\"\"<\/figure>\n\n\n\n

4\u30013D\u533b\u5b66\u56fe\u50cf\u5206\u5272\uff1a<\/p>\n\n\n\n

\"\"<\/figure>\n\n\n\n

\u4eca\u5929\uff0c\u5c31\u6765\u770b\u770bSwin Transformer \u5bf9\u4e8e\u56fe\u50cf\u53bb\u566a\u4efb\u52a1\u7684\u5904\u7406\u6548\u679c\uff1a<\/p>\n\n\n\n

\"\"<\/figure>\n\n\n\n
\"\"<\/figure>\n\n\n\n

\u4e2a\u4eba\u89c9\u5f97 Swin Transformer \u5bf9\u4e8e\u53bb\u566a\u6765\u8bf4\u8fd8\u6709\u5f88\u5927\u7684\u6269\u5c55\u7a7a\u95f4\uff0c\u8fd9\u7bc7\u8bba\u6587\u7684\u6a21\u578b\u6548\u679c\u4e0d\u662f\u5f88\u597d\uff0c\u53ef\u4ee5\u503c\u5f97\u53bb\u5c1d\u8bd5\u5c1d\u8bd5\uff0c\u770b\u770b\u6709\u6ca1\u6709\u66f4\u597d\u7684\u65b9\u6cd5\u63d0\u9ad8\u6a21\u578b\u6548\u679c\u3002<\/p>\n\n\n\n

\u8bba\u6587\u7684\u4e3b\u8981\u8d21\u732e\uff1a<\/p>\n\n\n\n

1\u3001\u7ed3\u5408Unet\u7f51\u7edc+ Swin Transformer <\/p>\n\n\n\n

2\u3001\u63d0\u51fa\u4e86\u4e00\u4e2a\u53cc\u4e0a\u91c7\u6837\u6a21\u5757 dual up-sample block<\/p>\n\n\n\n

3\u3001\u9996\u4e2a\u5c06Swin +unet\u7528\u4e8e\u56fe\u50cf\u53bb\u566a\u9886\u57df<\/p>\n\n\n\n

4\u3001\u5728 \u4e24\u4e2a\u901a\u7528\u6570\u636e\u96c6\u4e2d\u6d4b\u8bd5\u7684\u7ed3\u679c\u8fd8\u4e0d\u9519<\/p>\n\n\n\n

\u7f51\u7edc\u7ed3\u6784\uff1a<\/p>\n\n\n\n

\"\"<\/figure>\n\n\n\n

\u7f51\u7edc\u5206\u4e3a\u4e09\u4e2a\u90e8\u5206\uff1a1)Shallow feature extraction; 2) UNet feature extraction; and
3) Reconstruction module<\/p>\n\n\n\n

1\u3001Shallow feature extraction <\/h2>\n\n\n\n

\u4f7f\u75283*3\u5377\u79ef\uff0c\u63d0\u53d6\u7279\u5f81\uff0c\u8f93\u51fa\u901a\u905396<\/p>\n\n\n\n

2\u3001 UNet feature extraction <\/h2>\n\n\n\n

\u5e26\u6709 Swin Transformer Block \u7684UNET\u4f53\u7cfb\u7ed3\u6784\uff0c\u5176\u4e2d\u5305\u542b8\u4e2a Swin Transformer \u5c42\uff0c\u4ee5\u53d6\u4ee3\u5377\u79ef\u3002
Swin Transformer Block\uff08STB\uff09\u548cSwin Transformer\u5c42\uff08STL\uff09\uff1a<\/p>\n\n\n\n

\"\"
STB\uff1a\u5305\u542b8\u4e2aSTL<\/figcaption><\/figure><\/div>\n\n\n\n

\u8fd9\u5757\u5efa\u8bae\u53bb\u770b Swin Transformer \u8bba\u6587\uff0c\u8bb2\u7684\u6bd4\u8f83\u6e05\u695a\u3002\u6ce8\u610f\u6b64\u65f6\u7684\u8f93\u5165\u8f93\u51fa\u5927\u5c0f\u5b8c\u5168\u4e00\u81f4\uff0c\u56e0\u6b64\u9700\u8981\u4e0b\u91c7\u6837\u3002<\/p>\n\n\n\n

\u4e0b\u91c7\u6837\uff1a Patch merging <\/h2>\n\n\n\n

Patch merging:\u901a\u8fc7\u67e5\u770bPatch merging\u7684\u6e90\u7801\uff0c\u53ef\u4ee5\u770b\u5230\uff0c\u5176\u5b9e\u5c31\u662f\u4e00\u4e2a\u4e0b\u91c7\u6837\u7684\u8fc7\u7a0b\uff0c\u5b83\u53ef\u4ee5\u770b\u6210\u4e00\u79cd\u52a0\u6743\u6c60\u5316\u7684\u8fc7\u7a0b\u3002\u5b9e\u73b0\u7ef4\u5ea6\u4e0b\u91c7\u6837\u3001\u7279\u5f81\u52a0\u500d\u7684\u6548\u679c\u3002<\/p>\n\n\n\n

class PatchMerging(nn.Module):\n    def __init__(self, input_resolution, dim, norm_layer=nn.LayerNorm):\n        super().__init__()\n        self.input_resolution = input_resolution\n        self.dim = dim\n        self.reduction = nn.Linear(4 * dim, 2 * dim, bias=False)\n        self.norm = norm_layer(4 * dim)\n\n    def forward(self, x):\n        \"\"\"\n        x: B, H*W, C\n        \"\"\"\n        H, W = self.input_resolution\n        B, L, C = x.shape\n        assert L == H * W, \"input feature has wrong size\"\n        assert H % 2 == 0 and W % 2 == 0, f\"x size ({H}*{W}) are not even.\"\n\n        x = x.view(B, H, W, C)\n\n        x0 = x[:, 0::2, 0::2, :]  # B H\/2 W\/2 C<\/em>\n        x1 = x[:, 1::2, 0::2, :]  # B H\/2 W\/2 C<\/em>\n        x2 = x[:, 0::2, 1::2, :]  # B H\/2 W\/2 C<\/em>\n        x3 = x[:, 1::2, 1::2, :]  # B H\/2 W\/2 C<\/em>\n        x = torch.cat([x0, x1, x2, x3], -1)  # B H\/2 W\/2 4*C<\/em>\n        x = x.view(B, -1, 4 * C)  # B H\/2*W\/2 4*C<\/em>\n\n        x = self.norm(x)\n        x = self.reduction(x)\n\n        return x<\/code><\/pre>\n\n\n\n
\u4e0a\u91c7\u6837\uff1aDual up-sample<\/h2>\n\n\n\n
\u4f5c\u8005\u63d0\u51fa\u4e86 \u4e0a\u91c7\u6837\uff0c<\/p>\n\n\n\n
\u8be5\u6a21\u5757\u5305\u62ec\u4e24\u79cd\u73b0\u6709\u7684\u4e0a\u6837\u672c\u65b9\u6cd5\uff08\u5373\u53cc\u7ebf\u6027\u548cPixelShuffle\uff09\uff0c\u4ee5\u9632\u6b62\u68cb\u76d8\u4f2a\u5f71\uff08Deconvolution and Checkerboard Artifacts\u4e2d\u63d0\u51fa\u7684\uff09https:\/\/distill.pub\/2016\/deconv-checkerboard\/<\/a><\/em>  \u4ea7\u751f\u539f\u56e0\uff1a\u4e3b\u8981\u662f\u51fa\u73b0\u5728\u53cd\u5377\u79ef\u4e2d\u3002<\/p>\n\n\n\n
\"\"<\/figure>\n\n\n\n
\"\"
\u4e0a\u91c7\u6837\u6a21\u5757<\/figcaption><\/figure><\/div>\n\n\n\n
\u901a\u8fc7\u4e24\u79cd\u4e0a\u91c7\u6837\u540e,cat\u7ef4\u5ea6\u62fc\u63a5\u540e\uff0c\u901a\u8fc7\u4e00\u4e2a\u5377\u79ef\u5c42\u5c06\u7ef4\u5ea6\u51cf\u534aC\/2<\/p>\n\n\n\n
\u5b9e\u9a8c\uff1a<\/h2>\n\n\n\n
\u5982\u4e0a\u56fe\u6240\u793a\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"
ISCAS 2022\u7684\u4e00\u7bc7\u6587\u7ae0\uff0c\u4f5c\u4e3a\u9996\u4e2aSwin Transformer\u5728\u56fe\u50cf\u53bb\u566a\u9886\u57df\u7684\u5e94\u7528\uff0c\u6548\u679c\u6765\u8bf4\u611f\u89c9\u8fd8\u6709 … \u7ee7\u7eed\u9605\u8bfbSUNet: Swin Transformer with UNet for Image Denoising<\/span><\/a><\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":[],"categories":[18,9],"tags":[],"_links":{"self":[{"href":"http:\/\/139.9.1.231\/index.php\/wp-json\/wp\/v2\/posts\/6419"}],"collection":[{"href":"http:\/\/139.9.1.231\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"http:\/\/139.9.1.231\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"http:\/\/139.9.1.231\/index.php\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"http:\/\/139.9.1.231\/index.php\/wp-json\/wp\/v2\/comments?post=6419"}],"version-history":[{"count":44,"href":"http:\/\/139.9.1.231\/index.php\/wp-json\/wp\/v2\/posts\/6419\/revisions"}],"predecessor-version":[{"id":6474,"href":"http:\/\/139.9.1.231\/index.php\/wp-json\/wp\/v2\/posts\/6419\/revisions\/6474"}],"wp:attachment":[{"href":"http:\/\/139.9.1.231\/index.php\/wp-json\/wp\/v2\/media?parent=6419"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"http:\/\/139.9.1.231\/index.php\/wp-json\/wp\/v2\/categories?post=6419"},{"taxonomy":"post_tag","embeddable":true,"href":"http:\/\/139.9.1.231\/index.php\/wp-json\/wp\/v2\/tags?post=6419"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}