{"id":1707,"date":"2020-04-12T23:40:47","date_gmt":"2020-04-12T15:40:47","guid":{"rendered":"https:\/\/www.deepfaker.xyz\/?p=1707"},"modified":"2020-04-12T23:57:51","modified_gmt":"2020-04-12T15:57:51","slug":"dfl0325%e6%9b%b4%e6%96%b0%ef%bc%9axseg%e7%ae%80%e5%8c%96%e5%92%8c%e6%96%b0%e5%a2%9euhd%e7%bb%93%e6%9e%84%ef%bc%81","status":"publish","type":"post","link":"https:\/\/www.deepfaker.xyz\/?p=1707","title":{"rendered":"DFL0325\u66f4\u65b0\uff1aXseg\u7b80\u5316\u548c\u65b0\u589euhd\u7ed3\u6784\uff01"},"content":{"rendered":"

\u201c\u62fc\u547d\u4e09\u90ce\u201d\u53c8\u8fdb\u5165\u65e5\u66f4\u6a21\u5f0f\u4e86\uff1f\u521a\u4e0b\u5b8c24\u53f7\uff0c25\u7684\u7248\u672c\u53c8\u51fa\u73b0\u4e86\u3002\u4eba\u5bb6\u8fd9\u4e48\u52e4\u52b3\u6211\u4e5f\u4e0d\u80fd\u61d2\u60f0\u3002\u5148\u7ed9\u5927\u5bb6\u53d1\u4e2a\u66f4\u65b0\u901a\u77e5\u3002<\/p>\n

\u6211\u603b\u7ed3\u4e86\u4e0b\uff0c25\u53f7\u7248\u672c\u4e3b\u8981\u66f4\u65b0\u4e86\u4e24\u9879\u5185\u5bb9\u3002<\/p>\n

1. SAEHD\u6a21\u578b\u6dfb\u52a0\u4e86\u4e24\u79cd\u65b0\u7ed3\u6784\uff0c\u5206\u522b\u662fdfuhd\u548cliaeuhd\u3002<\/strong><\/p>\n

\u5173\u4e8e\u8fd9\u4e24\u79cd\u7ed3\u6784\u7684\u5b98\u65b9\u8bf4\u6cd5\u662f\uff1a<\/p>\n

\n
\n
uhd version is lighter than ‘HD’ but heavier than regular version.liaeuhd provides more “src-like” result<\/section>\n<\/section>\n<\/blockquote>\n

\u7ffb\u8bd1\u4e00\u4e0b\u5927\u6982\u5c31\u662f\uff1aUHD\u7248\u672c\u6bd4HD\u8981\u8f7b\uff08\u8f7b\u91cf\u7ea7\uff09\uff0c\u6bd4\u5e38\u89c4\u7248\u672c\u8981\u91cd\uff08\u91cd\u91cf\u7ea7\uff09\u3002liaeuhd\u53ef\u4ee5\u751f\u6210\u66f4\u65b0src\u7684\u7ed3\u679c\uff08\u5e94\u8be5\u662f\u76f8\u5bf9\u4e8eliae\u800c\u8a00\uff09\u3002<\/p>\n

 <\/p>\n

\u4e0b\u9762\u8bf7\u770b\u56fe\u7247\u5bf9\u6bd4<\/p>\n

\"\"<\/p>\n

liae<\/p>\n

\"\"<\/p>\n

liaeuhd<\/p>\n

\u56e0\u4e3aHD\uff0cUHD\u548cliae\u76ee\u524d\u7528\u7684\u6bd4\u8f83\u5c11\uff0c\u5c31\u4e0d\u591a\u8bf4\u4e86\uff0c\u6709\u65f6\u95f4\u6709\u8bbe\u5907\u7684\u53ef\u4ee5\u53bb\u7814\u7a76\u4e0b\u8fd9\u4e24\u4e2a\u7ed3\u6784\u3002<\/p>\n

2. \u7b80\u5316\u4e86Xseg\u7684\u6d41\u7a0b<\/strong><\/p>\n

\u524d\u51e0\u5929\u5199\u8fc7\u4e00\u7bc7\u300aDFL\u65b0\u529f\u80fd\u8be6\u89e3\uff0c\u4e00\u4e3e\u89e3\u51b3\u989d\u5934\u548c\u906e\u6321\u7684\u95ee\u9898\uff01\u300b<\/a>\u7684\u6587\u7ae0\uff0c\u4e3b\u8981\u4ecb\u7ecd\u7684\u5c31\u662fXseg\u7684\u4f7f\u7528\u5168\u8fc7\u7a0b\u3002<\/p>\n

\"\"<\/p>\n

\u5f53\u65f6\u6d89\u53ca\u5230Xseg\u7684\u811a\u672c\u5927\u6982\u67097\u4e2a\u3002<\/p>\n

\"\"<\/p>\n

 <\/p>\n

325\u7248\u672c\u4e2d\u7cbe\u7b80\u4e3a\u4e09\u4e2a\u811a\u672c\u3002<\/p>\n

\"\"<\/p>\n

Xseg\u7684\u5927\u7684\u4f7f\u7528\u6d41\u7a0b\u8fd8\u662f\u4e00\u6837\u7684\u3002\u5177\u4f53\u4f7f\u7528\u65b9\u6cd5\u53c2\u8003\u6211\u4e4b\u524d\u7684\u6587\u7ae0\uff0c\u53ef\u4ee5\u7acb\u9a6c\u5c31\u7528\u8d77\u6765\u4e86\u3002<\/p>\n

\"\"<\/p>\n

 <\/p>\n

\u8fd8\u6709\u4e00\u70b9\u5c31\u662f\uff0c325\u7684Xseg\u7f16\u8f91\u5668\u53d8\u4e86\uff0c\u64cd\u4f5c\u548c315\u8fd8\u6709\u70b9\u4e0d\u4e00\u6837\u3002\u66f4\u65b0\u592a\u5feb\uff0c\u5199\u6587\u7ae0\u7684\u901f\u5ea6\u66f4\u4e0d\u4e0a\u4eba\u5bb6\u5199\u4ee3\u7801\u901f\u5ea6…315\u5df2\u7ecf\u5b66\u4f1a\u7684\u5c31\u7528315\u3002\u5176\u5b9e\uff0c2019\u5e74\u7684620\u4e5f\u4e0d\u5dee\u554a\uff0c\u53ea\u8981\u9002\u7528\u4e8e\u81ea\u5df1\u7684\u5e94\u7528\u573a\u666f\u5373\u53ef\u3002<\/p>\n

===========================<\/p>\n

\u5177\u4f53\u66f4\u65b0\u65e5\u5fd7\uff1a<\/p>\n

SAEHD: added ‘dfuhd’ and ‘liaeuhd’ archi<\/p>\n

SAEHD: \u6dfb\u52a0\u4e86dfuhd\u548cliaeuhd\u7ed3\u6784<\/p>\n

uhd version is lighter than ‘HD’ but heavier than regular version.<\/p>\n

uhd\u7248\u672c\u6bd4HD\u66f4\u52a0\u8f7b\u91cf\u7ea7\uff0c\u6bd4\u5e38\u89c4\u7248\u672c\u66f4\u52a0\u91cd\u91cf\u7ea7\u3002<\/p>\n

liaeuhd provides more “src-like” result<\/p>\n

liaeuhd\u63d0\u4f9b\u4e86\u66f4\u52a0\u50cfsrc\u7684\u7ed3\u679c<\/p>\n

comparison:<\/p>\n

\u6bd4\u8f83<\/p>\n

liae: \u00a0 \u00a0https:\/\/i.imgur.com\/JEICFwI.jpg<\/p>\n

liaeuhd: https:\/\/i.imgur.com\/ymU7t5E.jpg<\/p>\n

added new XSegEditor !<\/p>\n

\u6dfb\u52a0\u65b0\u7684Xseg\u7f16\u8f91\u5668\u3002<\/p>\n

here new whole_face + XSeg workflow:<\/p>\n

\u4ee5\u4e0b\u4e3a\u6574\u8138\u548cXseg\u7684\u5de5\u4f5c\u6d41\u7a0b\u3002<\/p>\n

with XSeg model you can train your own mask segmentator for dst(and\/or src) faces<\/p>\n

\u4f7f\u7528Xseg\u6a21\u578b\uff0c\u4f60\u53ef\u4ee5\u81ea\u5df1\u4e3aDst\u8138\u90e8\u8bad\u7ec3\u906e\u7f69\u5206\u5272\u5668<\/p>\n

that will be used by the merger for whole_face.<\/p>\n

\u5408\u6210\u7684\u65f6\u5019\u4f1a\u5e94\u7528\u3002<\/p>\n

Instead of using a pretrained segmentator model (which does not exist),<\/p>\n

\u66ff\u4ee3\u4e86\u4f7f\u7528\u9884\u8bad\u7ec3\u906e\u7f69\u6a21\u578b\u7684\u65b9\u5f0f\uff08\u4e0d\u66fe\u5b58\u5728\u8fc7\u7684\u90a3\u79cd\u65b9\u5f0f\uff0c\u6b64\u5904\u662f\u5e7d\u9ed8\uff1f\uff09<\/p>\n

you control which part of faces should be masked.<\/p>\n

\u4f60\u53ef\u4ee5\u63a7\u5236\u54ea\u4e00\u90e8\u5206\u5e94\u8be5\u88ab\u906e\u6321\uff08\u5f53\u7136\uff0c\u5de5\u4f5c\u91cf\u4e5f\u5927\u4e86\u4e0d\u5c11\u3002\uff09<\/p>\n

new scripts:<\/p>\n

\u65b0\u7684\u811a\u672c\u3002<\/p>\n

5.XSeg) data_dst edit masks.bat<\/p>\n

5.XSeg) data_src edit masks.bat<\/p>\n

5.XSeg) train.bat<\/p>\n

Usage:<\/p>\n

\u4f7f\u7528\uff1a<\/p>\n

unpack dst faceset if packed<\/p>\n

\u6570\u636e\u96c6\u5982\u679c\u6253\u5305\u4e86\uff0c\u8bf7\u5148\u89e3\u538b\u3002<\/p>\n

run 5.XSeg) data_dst edit masks.bat<\/p>\n

\u8fd0\u884c\u811a\u672c<\/p>\n

Read tooltips on the buttons (en\/ru\/zn languages are supported)<\/p>\n

\u9605\u8bfb\u63d0\u793a\uff08\u652f\u6301\u82f1\u6587\uff0c\u4fc4\u8bed\uff0c\u4e2d\u6587\uff09<\/p>\n

mask the face using include or exclude polygon mode.<\/p>\n

\u4f7f\u7528\u5305\u542b\u548c\u6392\u9664\u7684\u65b9\u5f0f\u6807\u6ce8\u8138\u90e8\u3002<\/p>\n

repeat for 50\/100 faces,<\/p>\n

\u91cd\u590d 50\u5230100\u5f20\u8138<\/p>\n

!!! you don’t need to mask every frame of dst<\/p>\n

\u65e0\u9700\u6807\u6ce8\u6240\u6709\u5e27<\/p>\n

only frames where the face is different significantly,<\/p>\n

\u53ea\u8981\u5173\u6ce8\u90a3\u4e9b\u6709\u5dee\u5f02\u7684\u8138\u3002<\/p>\n

for example:<\/p>\n

\u4e3e\u4f8b<\/p>\n

closed eyes<\/p>\n

\u95ed\u773c<\/p>\n

changed head direction<\/p>\n

\u6539\u53d8\u8138\u7684\u65b9\u5411<\/p>\n

changed light<\/p>\n

\u6539\u53d8\u706f\u5149<\/p>\n

the more various faces you mask, the more quality you will get<\/p>\n

\u6807\u6ce8\u8d8a\u591a\uff0c\u6548\u679c\u8d8a\u597d\u3002<\/p>\n

Start masking from the upper left area and follow the clockwise direction.<\/p>\n

Keep the same logic of masking for all frames, for example:<\/p>\n

the same approximated jaw line of the side faces, where the jaw is not visible<\/p>\n

the same hair line<\/p>\n

Mask the obstructions using exclude polygon mode.<\/p>\n

run XSeg) train.bat<\/p>\n

\u8fd0\u884c\u811a\u672c<\/p>\n

train the model<\/p>\n

\u8bad\u7ec3\u6a21\u578b<\/p>\n

Check the faces of ‘XSeg dst faces’ preview.<\/p>\n

\u67e5\u770b\u9884\u89c8\u56fe<\/p>\n

if some faces have wrong or glitchy mask, then repeat steps:<\/p>\n

\u5982\u679c\u53d1\u73b0\u95ee\u9898\uff0c\u7ee7\u7eed\u4ee5\u4e0b\u6b65\u9aa4<\/p>\n

run edit<\/p>\n

find these glitchy faces and mask them<\/p>\n

train further or restart training from scratch<\/p>\n

Restart training of XSeg model is only possible by deleting all ‘model\\XSeg_*’ files.<\/p>\n

\u53ea\u6709\u5220\u9664\u4e86model\\XSeg-*\u7684\u6587\u4ef6\u624d\u80fd\u91cd\u65b0\u8bad\u7ec3\u3002<\/p>\n

If you want to get the mask of the predicted face (XSeg-prd mode) in merger,<\/p>\n

\u5982\u679c\u9700\u8981\u4f7f\u7528predicted face\u906e\u7f69<\/p>\n

you should repeat the same steps for src faceset.<\/p>\n

\u4f60\u9700\u8981\u5bf9src\u6570\u636e\u96c6\u91cd\u590d\u4e00\u6837\u7684\u6b65\u9aa4\u3002<\/p>\n

New mask modes available in merger for whole_face:<\/p>\n

\u8fd9\u4e2a\u906e\u7f69\u5728\u5408\u6210\u7684\u65f6\u5019\u88ab\u5e94\u7528\uff0c\u4ec5\u9488\u5bf9\u6574\u8138\u3002\u5177\u4f53\u9009\u9879\u5982\u4e0b\uff1a<\/p>\n

XSeg-prd \u00a0 \u00a0 \u00a0– XSeg mask of predicted face \u00a0 \u00a0-> faces from src faceset should be labeled<\/p>\n

XSeg-dst \u00a0 \u00a0 \u00a0– XSeg mask of dst face \u00a0 \u00a0 \u00a0 \u00a0 \u00a0-> faces from dst faceset should be labeled<\/p>\n

XSeg-prd*XSeg-dst – the smallest area of both<\/p>\n

if workspace\\model folder contains trained XSeg model, then merger will use it,<\/p>\n

\u5982\u679cmodel\u76ee\u5f55\u4e0b\u5305\u542bXseg\u6a21\u578b\uff0c\u5219\u88ab\u5e94\u7528\u3002<\/p>\n

otherwise you will get transparent mask by using XSeg-* modes.<\/p>\n

\u5426\u5219\uff0c\u60a8\u5c06\u901a\u8fc7\u4f7f\u7528XSeg-*\u6a21\u5f0f\u83b7\u5f97\u900f\u660e\u63a9\u7801\u3002<\/p>\n

Some screenshots:<\/p>\n

\u4e00\u4e9b\u622a\u56fe<\/p>\n

\"\"<\/p>\n

\"\"<\/p>\n

XSegEditor: https:\/\/i.imgur.com\/7Bk4RRV.jpg<\/p>\n

trainer \u00a0 : https:\/\/i.imgur.com\/NM1Kn3s.jpg<\/p>\n

merger \u00a0 \u00a0: https:\/\/i.imgur.com\/glUzFQ8.jpg<\/p>\n

example of the fake using 13 segmented dst faces<\/p>\n

\u4e00\u4e2a\u4f7f\u7528segmented\u7684\u52a8\u6001\u56fe<\/p>\n

: https:\/\/i.imgur.com\/wmvyizU.gifv<\/p>\n","protected":false},"excerpt":{"rendered":"

\u201c\u62fc\u547d\u4e09\u90ce\u201d\u53c8\u8fdb\u5165\u65e5\u66f4\u6a21\u5f0f\u4e86\uff1f\u521a\u4e0b\u5b8c24\u53f7\uff0c25\u7684\u7248\u672c\u53c8\u51fa\u73b0\u4e86\u3002\u4eba\u5bb6\u8fd9\u4e48\u52e4\u52b3\u6211\u4e5f\u4e0d\u80fd\u61d2\u60f0\u3002\u5148\u7ed9\u5927\u5bb6\u53d1\u4e2a\u66f4\u65b0\u901a\u77e5 […]<\/p>\n","protected":false},"author":1,"featured_media":1713,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":[],"categories":[3],"tags":[],"_links":{"self":[{"href":"https:\/\/www.deepfaker.xyz\/index.php?rest_route=\/wp\/v2\/posts\/1707"}],"collection":[{"href":"https:\/\/www.deepfaker.xyz\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.deepfaker.xyz\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.deepfaker.xyz\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.deepfaker.xyz\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=1707"}],"version-history":[{"count":3,"href":"https:\/\/www.deepfaker.xyz\/index.php?rest_route=\/wp\/v2\/posts\/1707\/revisions"}],"predecessor-version":[{"id":1794,"href":"https:\/\/www.deepfaker.xyz\/index.php?rest_route=\/wp\/v2\/posts\/1707\/revisions\/1794"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.deepfaker.xyz\/index.php?rest_route=\/wp\/v2\/media\/1713"}],"wp:attachment":[{"href":"https:\/\/www.deepfaker.xyz\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=1707"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.deepfaker.xyz\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=1707"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.deepfaker.xyz\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=1707"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}