routineLife1 / AFI-ForwardDeduplicate

Efficient Deduplicate for Anime Video Frame Interpolation
MIT License
5 stars 0 forks source link

Increase Processing Speed Possible? #5

Closed 2994dd42d75b0 closed 5 months ago

2994dd42d75b0 commented 5 months ago

Using GPU to covert and combine frames ffmpeg -i in.mkv -crf 16 -r 24000/1001-preset slow -c:v hevc_nvenc -profile:v main10 -c:a copy out.mkv

ffmpeg -r 60 -thread_queue_size 5120 -i /O/%09d.png -i out.mkv -map 0:v -map 1:a -map 1:s -preset slow -gpu 0 -c:v hevc_nvenc -profile:v main10 -c:a copy output.mkv

Parallelize Frame Reading: Use multiple threads to read and resize frames concurrently.

Use PyTorch DataLoader: Utilize torch.utils.data.DataLoader for batch processing and to handle data loading efficiently.

Batch Inference: Ensure inference is performed on batches of frames instead of one frame at a time.

Can this increase speed,My vram is not fully used Rife is fast but gmfss is very slow

routineLife1 commented 5 months ago

Thank you for paying attention to our work and making suggestions. We have tried all the methods you said, and it cannot be significantly accelerated significantly. A more cost-effective approach is to wait for faster and more advanced vfi algorithms to instead

2994dd42d75b0 commented 5 months ago

Thanks for reply I'll wait for gmfss.