University of Maryland and Adobe Research collaborated to create a groundbreaking VideoGigaGAN model capable of upscaling videos up to 8 times while maintaining rich details and temporal consistency. By building upon the successful image upsampler GigaGAN, VideoGigaGAN addresses key issues such as temporal flickering and aliasing artifacts to produce high-quality, detail-rich results. The model outperforms previous Video Super-Resolution approaches and showcases impressive results on public datasets. By incorporating innovative techniques like temporal attention layers and flow-guided propagation modules, VideoGigaGAN sets a new standard for video upscaling technology. (Word count: 100)
https://videogigagan.github.io/