Forked from OpenMMLab --> this repo will automaticly update when original aurthers make changes!
Great for GPU accelarated Optical Flow calculations
- Demo:
python demo/video_demo.py demo/demo.mp4 \
configs/raft/raft_8x2_100k_mixed_368x768.py \
checkpoints/raft_8x2_100k_mixed_368x768.pth \
raft_demo.mp4 --gt demo/demo_gt.mp4
- on custome data/videos with raft model:
python demo/video_demo.py {video_input.mp4} \
configs/raft/raft_8x2_100k_mixed_368x768.py \
checkpoints/raft_8x2_100k_mixed_368x768.pth \
{output_fn.mp4}
- more general:
python demo/video_demo.py {video_input}.mp4 \
configs/{model_class}/{model_name}.py \
checkpoints/{model_name}.pth \
results/{output_fn).mp4
This algorithm takes lots of Memory in both RAM and GPU!
Try to split input videos in to smaller pieces, so that they can fit in the limited resources!
A 5 minutes video could fit in perfect in my setting:
- 10900X
- 32GB * 2 RAM
- RTX 3080Ti 12GB * 2 --> not sure if I can use both for the same task