search

OpenVisualCloud / Video-Super-Resolution-Library.Enhanced-BasicVSR

Video super resolution (VSR) is used in AI media enhancement to convert low-resolution video to high-resolution. BasicVSR is a public AI algorithm. We have enhanced the public model to achieve better visual quality and less computational complexity. The performance of BasicVSR inference has also been optimized for Intel Data Center GPU Flex 170.
BSD 3-Clause "New" or "Revised" License
10 stars 2 forks source link

question about pipeline? #1

Open Feynman1999 opened 1 year ago

Feynman1999 commented 1 year ago

Based on my poor knowledge, my understanding is that ffmpeg filters only support frame by frame synchronized image processing? And BasicVSR requires many frames before and after, can you explain how you support it? It would be even better if there was an overall technical introduction to help me understand this project.

djie1 commented 7 months ago

it queues several frames for inference in the filter. Implementation details in https://github.com/OpenVisualCloud/Video-Super-Resolution-Library.Enhanced-BasicVSR/blob/master/ivsr_ffmpeg_plugin/patches/0002-Enable-multi-frame-fifo-with-BasicVSR-model.patch