Produced by: Tarun Mishra
Adobe Research's team of video and AI engineers has introduced VideoGigaGAN, an AI application designed to enhance the sharpness and clarity of blurry videos, representing a significant leap in video upscaling technology.
The team has detailed its findings and methodology in an article published on the arXiv preprint server, offering insights into the development and capabilities of VideoGigaGAN.
VideoGigaGAN has utilized generative adversarial networks (GANs), a form of AI architecture, to discern and replicate the features of sharp and clear video imagery.
To ensure consistency across video frames, a "flow-guided propagation module" has been integrated into VideoGigaGAN, enhancing the coherence of the upscaled output.
The application has incorporated anti-aliasing techniques and high-frequency feature shuttling to address common issues associated with AI-generated content, such as unnatural colouring and uneven lines.
Adobe's team asserts that VideoGigaGAN can upscale video image quality by up to eight times, achieving notable improvements without introducing noticeable distortions or aberrations.
Acknowledging the system's reliance on estimations for missing imagery, the team has highlighted the addition of fine details like skin pores and eyelashes to enhance overall video quality.
While the unveiling of VideoGigaGAN has showcased its promising capabilities, its release for general use remains uncertain, with Adobe yet to confirm its availability beyond the demonstration phase.