rife-ncnn-vulkan

rife-ncnn-vulkan copied to clipboard

Did NCNN version impact the result?

3

I build the latest ncnn on win10, and run extract the cpu-forward code from this github. But when I try to test it with the two images from this github,...

GuideWsp

Please provide script to interpolate video in one line

4

I've written a template python script. ```python import sys import os import re import shutil if len(sys.argv) < 2: print("Usage: python3 start.py ") exit(1) video = sys.argv[1] if not os.path.exists(video):...

yusing

set up rife-ncnn-vulkan on google colab

2

It would be awesome if anyone could build on google colab for me! Because I've been getting this error: ``` /content/rife-ncnn-vulkan-20220330-ubuntu/rife-ncnn-vulkan: error while loading shared libraries: libvulkan.so.1: cannot open shared...

mikebilly

Using the same Rife model 2.3, rife-ncnn-vulkan creates outputs not as smooth as rife-cuda. However, rife-cuda has more warping between the frames than rife-ncnn-vulkan. Why is this?

mikebilly

Timestep argument does not work at all

1

Using `-s` has no effect (RIFE v4). Timestep always defaults to 0.5.

n00mkrad

add modelv4.1

1

看了一下v4.1的网络结构似乎和v4没区别, 于是尝试转了一下4.1的模型 param用的还是v4的param, 只是更新了权重 附上测试图(由README的Sample Images缩放到448x256进行处理) ncnn-v4.1  ncnn-v4  官方-v4.1  ❤️ 

mirrorsysu

binding_count not match, expect 9 but got 2 + 0 in rife-v4 model. (rife-v4 model does not support TTA.)

2

I tried to utilize the v4 model, but that error causes problems. Only v4 models are missing `contextnet.bin`, `contextnet.param` and `fusionnet.bin`, `fusionnet.param` files. Is this missing a model that is...

Kafu-Yukimura

RIFE built from source stops processing after a few images.

1

KCGD

GPU generates black image on Adreno 618/660 (Android/Termux)

1

Thanks a lot for porting this RIFE to ncnn! I recently tried to use this project on Termux/Android. `rife-ncnn-vulkan` generates black image using GPU while perfect running CPU (`rife-ncnn-vulkan -g...

knyipab

ncnn-vulkan time performance

1

您好。我用了这个ncnn 的推理框架，在 nvidia-T4上测试发现，它相比于 pytorch 实现的rife， 速度慢了很了。推理时间几乎达到了pytorch 版本的 3-5倍左右。请问这个原因是什么呀？是不是在云上ncnn 的框架相比于pytorch 没有速度优势呢？

liye0713