nihui comments

Results 477 comments of


                                            nihui

SSR-Net keras -> onnx -> ncnn, Segmentation fault (core dumped) error.

Hi, try keras2ncnn https://github.com/MarsTechHAN/keras2ncnn

vulkan和cuda速度对比

nvidia卡用cuda速度更快，因为cuda是nvidia自己对着自己硬件优化的 vulkan是通用，优势在于更广泛的兼容性，更小的二进制包，更少的显存占用这些

> 目前我手上的卡的速度 realcugan, 1080p-2160p RTX3090, 183f/min Tesla P40, 62f/min Quadro M1200, 14f/min RX480, 10f/min > > realesrgan xs, 1080p-2160p RTX3090, 220f/min Tesla P40, 80f/min RX480, 34f/min Quadro M1200, 15f/min 试试...

vulkan和cuda速度对比

> > > 目前我手上的卡的速度 realcugan, 1080p-2160p RTX3090, 183f/min Tesla P40, 62f/min Quadro M1200, 14f/min RX480, 10f/min > > > realesrgan xs, 1080p-2160p RTX3090, 220f/min Tesla P40, 80f/min RX480, 34f/min Quadro...

Implement SPIR-V archives for combining shader modules.

afaik, multiple entry point in one shader module did not work with moltenvk when I tried this approach two years ago. so in ncnn library, the one shader one entry...

Is it possible to create standalone program for that?

you can try waifu2x-ncnn-vulkan, works on almost all GPU https://github.com/nihui/waifu2x-ncnn-vulkan

用darknet框架来做这个任务的优势在哪儿

大佬你好，我是你的粉丝！

From where and how did you convert the squeezenet model to ncnn?

caffe squeezenet

From where and how did you convert the squeezenet model to ncnn?

https://github.com/Tencent/ncnn/wiki/use-ncnn-with-alexnet

Use ncnn low level api instead of eigen for float convolutions

daquexian is awesome !