之前:《简单介绍SimSwap(类似DeepFaceLab)单张图视频换脸的项目》
采用的分辨率是224。
最近发现有更新:
About using Simswap 512 (beta version)
We trained a beta version of Simswap 512 on VGGFace2-HQ and open sourced the model (if you think the Simswap 512 is cool, please star our VGGFace2-HQ repo).
The usage of applying Simswap 512 is to modify the value of the argument: “–crop_size” to 512
简单说就是把命令行里面的:
--crop_size 224
改为:
--crop_size 512
为此你需要先拉取项目(git clone git@github.com:neuralchen/SimSwap.git)最新的代码。
再下载512.zip,并解压放置到./checkpoints目录:
Simswap 512 (optional)
The checkpoint of Simswap 512 beta version has been uploaded in Github release.If you want to experience Simswap 512, feel free to try.
Unzip 512.zip, place it in the root dir ./checkpoints.
确认你的GPU至少有3GB显存。
再执行类似指令:
python test_wholeimage_swapmulti.py --crop_size 512 --use_mask --Arc_path arcface_model/arcface_checkpoint.tar --pic_a_path ./demo_file/faceIn.JPG --pic_b_path ./demo_file/faceOut.jpg --output_path ./output/
试了一下,分辨率明显高不少。
但和想象中不太一样,
和官网例子也似乎不同。。。
大概对比如下图:
呃…… the end。