PeterL1n / RobustVideoMatting

Robust Video Matting in PyTorch, TensorFlow, TensorFlow.js, ONNX, CoreML!
https://peterl1n.github.io/RobustVideoMatting/
GNU General Public License v3.0
8.63k stars 1.14k forks source link

4090 #232

Open twkim37 opened 1 year ago

twkim37 commented 1 year ago

(rvm) I:\RVM\RobustVideoMatting-master>python inference.py --variant mobilenetv3 --checkpoint checkpoint/rvm_mobilenetv3.pth --device cuda --input-source "input/input.mp4" --downsample-ratio 0.25 --output-type video --output-composition "output/composition.mp4" --output-alpha "output/alpha.mp4" --output-foreground "output/foreground.mp4" --output-video-mbps 8 --seq-chunk 12 47%|██████████████████████████████████▍ | 7092/15059 [1:07:44<1:16:21, 1.74it/s]

...

I have a 4090 and my GPU utilization is significantly lower.

I use CUDNN version 11.8. I tried 11.7 just in case, but the speed is the same.

The GPU load in the task manager is only 2%. Sometimes it goes to 20%, but only briefly. My gpu power usage with hwmonitor is also only 45W.

The GPU is not being utilized. Is there something I'm missing?

JuggerGuGu commented 1 year ago

In the task manager, there are different types of GPU utilization, such as 3D, Copy, Video Encode, and more. Make sure you have switched to the "CUDA" option.