This update adds SAEHD: lr_dropout parameters, you can open or disable (disabled by default) training, each face transplant after adequate training can enable this option to reduce the number of repetitions to obtain additional clarity . There is also a more meaningful update is to increase the picture metadata function, simply means that you use other software (such as Photoshop and Topaz Gigapixel AI (https://www.deepfacelabs.com/read-64-1.html ) Figure modified face after a simple conversion can also be used directly)
That is equal to update this default so that small partners with software to enhance clarity of the face map to achieve the best training effect. However, please note: only clarity, no matter how much you cut a face chart, authors or let you convert back to face a 256x256 map, not to pursue a face view of 512 or 1024, the final software-to-face view of an input or 256PX , large the resolution meaningless, do not do useful work.
添加：4.2.other) data_src util faceset metadata save.bat
将data_src \ aligned \ faces的元数据保存到data_src \ aligned \ meta.dat中。
添加：4.2.other) data_src util faceset metadata restore.bat
将元数据从“ meta.dat”还原到图像，如果图像尺寸与原始尺寸不同，则会自动调整尺寸。
使用方法：
1、下载安装Topaz Gigapixel AI(https://www.deepfacelabs.com/read-64-1.html)软件。
2、运行4.2.other) data_src util faceset metadata save.bat
3、打开Topaz Gigapixel AI
4、用Topaz Gigapixel AI打开你电脑中data_src\aligned所有需要增强的脸图
5、将输出文件夹设置为“ data_src \ aligned_topaz”（在保存对话框中创建文件夹）
6、将设置设为图二中的选项，可以选择2倍，4倍或6倍的放大率。
7, to start the conversion
8, rename a folder:
data_src \ aligned-> data_src \ aligned_original
data_src \ aligned_topaz-> data_src \ the aligned
. 9, copy 'data_src \ aligned_original \ meta.dat' to 'data_src \ the aligned \'
10, running data_src util faceset metadata restore.bat, this time will be reduced to the original image size (256x256), while maintains detailed information metadata.
11. You can now use the new enhanced face up!
Download Link:
Google Cloud disk:
Links: https://drive.google.com/drive/folders/1BCFK_L7lPNwMbEQ_kFPqPpDdFEOd_Dci
Baidu network disk:
Links: https://pan.baidu.com/s/1KywG-K1vrHHkLT6NZsGb9w
Extraction code:
vgye