search

nihui / dain-ncnn-vulkan

DAIN, Depth-Aware Video Frame Interpolation implemented with ncnn library
MIT License
515 stars 45 forks source link

Creates black images with "vkQueueSubmit failed -4" #19

Open fluffyloopy opened 4 years ago

fluffyloopy commented 4 years ago

image

PumpkinCheshire commented 4 years ago

I think it's a bug of graph card. I have same error with intel, but fine with Nvidia card. I guess it may be related to the Graphics RAM or bit width. Maybe reducing the GPU usage of other programs will help. Just guessing irresponsibly, I don’t understand this field.

fluffyloopy commented 4 years ago

Doesnt work with my Nvidia and i dont know how to switch to intel. Checked through task manager and it shows 100% being used on Nvidia GPU. It usually has 2% being used. 2GB Vram.

nihui commented 3 years ago

try smaller tile-size btw, you may interest in https://github.com/nihui/cain-ncnn-vulkan or https://github.com/nihui/rife-ncnn-vulkan

fluffyloopy commented 3 years ago

Using smaller tile-size leads to this image RIFE and CAIN works perfectly

Update : rife and cain works for some stuff, produces black or messed up images for others same as dain while dain doesnt work on anything

Hurous commented 3 years ago

image When you use dain-ncnn-vulkan, you can set -g gpu-id to switch gpu

usernamecannotbeempty commented 2 years ago

I encountered the same problem even with rife-ncnn-vulkan Windows 10, i7-6700, 8GB RAM, GTX 1050 Ti 4GB VRAM. I shutdown all other programs that use VRAM. So, I am not convinced it's due to insufficient GPU VRAM at all.

As expected, running the interpolation again only with the previous and the next image is totally fine. Something is wrong when rife-ncnn-vulkan running in batch mode. I then wrote a small python code to automatically repeat this action. fix_black_frames.py.txt