waifu2x-ncnn-vulkan
waifu2x-ncnn-vulkan copied to clipboard
nihui
waifu2x converter ncnn version, runs fast on intel / amd / nvidia / apple-silicon GPU with vulkan
Right now (as of b152fc7ee703ea46575fb0d20fef18cb7237236c), waifu2x-ncnn-vulkan reports exit code 0 (on Linux) no matter what, even if there's an error. This breaks scripting, for example when trying to work around...
我发现所有的ncnn-vulkan超分辨率项目(waifu2x,srmd,realsr)均不能正常保留下面这个图片的透明通道:  我个人怀疑是在检测图片是否存在alpha channel上出了问题. 其他软件在处理这个图片时均可检测并保留这个图片的alpha通道.
Currently there seems to be no way to hide all of this: ``` [0 AMD RADV/ACO VEGA10 (LLVM 10.0.1)] queueC=1[4] queueG=0[1] queueT=0[1] [0 AMD RADV/ACO VEGA10 (LLVM 10.0.1)] bugsbn1=0 buglbia=0...
is it possible to upscale a directory while making waifu2x skip some files if they have the same name in both the input and output directories?
很抱歉英文不好，只能用中文写。我的意思是假如在转换大量图片过程中因为某些原因中断的话。目前就只能全部重新转换（因为转换顺序好像是随机的，所以无法筛选出已转换文件），所以希望能够加上类似的选项以节省时间。
So we could get a bit faster processing, and embedding within our apps... Thanks!
Show difference quality as a numerical value between original (full size) image and upscaled image in the waif2x implementation comparison chart. Probably using [SSIM](https://en.wikipedia.org/wiki/Structural_similarity). Would be a informative comparison.
waifu2x-caffe seems to support some robust resolution output options as seen [here](https://github.com/lltcggie/waifu2x-caffe#%E6%8B%A1%E5%A4%A7%E3%82%B5%E3%82%A4%E3%82%BA), whereas vulkan only supports 2x. Just wondering. Thanks.
No output image for this image in AlmaLinux 8.6 https://cdn.discordapp.com/attachments/352403408264757248/1021481890525483009/cover_test.webp in PHP `echo shell_exec('waifu2x-ncnn-vulkan -i "'.$source.'" -o "'.$output.'" -m models-upconv_7_anime_style_art_rgb 2>&1; echo $?');` I get this output `vkEnumeratePhysicalDevices failed -3...
