realcugan-ncnn-vulkan
realcugan-ncnn-vulkan copied to clipboard
nihui
real-cugan converter ncnn version, runs fast on intel / amd / nvidia / apple-silicon GPU with vulkan
Real-GUGAN has introduced a parameter called "alpha" in 20220227. > alpha: 该值越大AI修复程度、痕迹越小，越模糊；alpha越小处理越烈，越锐化，色偏（对比度、饱和度增强）越大；默认为1不调整，推荐调整区间(0.7,1.3)； It is possible to specify this parameter in realcugan-ncnn-vulkan? Is there any plan to support it? Thanks.
_wfopen H:\Tools\realcugan-ncnn-vulkan-20220318-windows\models-se/up4x-denoise1x.param failed Is what I get when trying to run 4x upscale against a folder of pngs. Same folder works in 2x upscale, 3x fails same as 4x. Command...
开四倍报错
运行： realcugan-ncnn-vulkan -i $file -o $file -s 4 -n 1 -x [0 Apple M1 Max] queueC=0[1] queueG=0[1] queueT=0[1] [0 Apple M1 Max] bugsbn1=0 bugbilz=188 bugcopc=0 bugihfa=0 [0 Apple M1 Max]...
Hi there, I'm trying to install module realcugan-ncnn-vulkan-python on my macbook, but getting an error: > aise CalledProcessError(retcode, cmd) > subprocess.CalledProcessError: Command '['cmake', '-S', '/private/var/folders/p9/mx7w4h2149vdglfr2c0ddmpm0000gn/T/pip-install-p6r_9mxi/realcugan-ncnn-vulkan-python_71a0be6d527745c59fa316681cfad201/realcugan_ncnn_vulkan_python', '-B', '/private/var/folders/p9/mx7w4h2149vdglfr2c0ddmpm0000gn/T/pip-install-p6r_9mxi/realcugan-ncnn-vulkan-python_71a0be6d527745c59fa316681cfad201/build/temp.macosx-10.9-x86_64-cpython-39_realcugan-ncnn-vulkan-python', '-GNinja', '-DCMAKE_BUILD_TYPE=Release', '-DCMAKE_INSTALL_PREFIX:PATH=/private/var/folders/p9/mx7w4h2149vdglfr2c0ddmpm0000gn/T/pip-install-p6r_9mxi/realcugan-ncnn-vulkan-python_71a0be6d527745c59fa316681cfad201/build/lib.macosx-10.9-x86_64-cpython-39/realcugan_ncnn_vulkan_python',...
realcugan-ncnn-vulkan:  realcugan:  Both upscaled the same source image 4x (2 times 2x). I used tta in realcugan-ncnn-vulkan.
``` image unkind\illust_96146558_20220214_121642.png has alpha channel ! unkind\illust_96146558_20220214_121642.png will output unkind\illust_96146558_20220214_121642.png0RCG1.5.jpg.png gap sync gap0 gap sync gap1 gap sync gap2 gap sync gap3 ``` 当出现 alpha channel 时，卡在这个进度。然后输出文件尺寸持续变大
 已确认是-x参数引起，不添加-x后无此问题
你好 首先非常感谢提供ncnn版本 想请教关于syncgap-mode的用途是什么 当放大解析度较大的图片时，syncgap-mode使用default会出现如下的错误，除此之外皆可正常使用，这是否是显存不足的问题  第二个问题是，使用no-sync会出现以下的长方形边界，是否不该提供no-sync的方式 [图源1](https://www.pixiv.net/artworks/95008985) [图源2](https://www.pixiv.net/artworks/95739807)   第三个是使用cugan后会有色偏的问题 [图源](https://www.pixiv.net/artworks/93387702) 原图  使用CUGAN后  [图源](https://twitter.com/u_m_o_u/status/1459200078162190337) 原图  使用CUGAN后 
请问一下，使用ncnn vulkan可以支持AMD卡，但是想问一下如果使用Nvidia显卡，那么ncnn vulkan和cuda的速度相比怎么样呢？ncnn vulkan会不会慢很多？谢谢。
