VidPanos - Google AI stabilises video and adds the missing edges

[09:23 Sun,3.November 2024 by Rudi Schmidts]

At the weekend, some members of the Google DeepMind research group, among others, published what we consider to be a highly interesting AI application and associated project website.

‘VidPanos’ (the working title) synthesises a panoramic video from a random, hand-captured panning video - as if the original video had been taken with a wide-angle camera. It works like a combination of a digital image stabiliser and generative fill for the missing edges of the video.

In fact, such an application should quickly find many users, as it can utilise the fact that human perception does not usually pay much attention to the generated peripheral areas. If a detail is not one hundred per cent correct, this is usually only noticed after repeated viewing. Put simply, VidPanos’ simply “invents” everything around the filmed video using a video AI diffusion model and stabilises it at the same time. The user does not have to make any further settings for this.

VidPanos - An AI stabilises the video and adds the missing border areas.

On the other hand, the benefits of such a tool are enormous for many applications. Quick and dirty tracking shots can be produced with practically no technical effort. In the worst-case scenario, the cameraman simply holds it somehow. At best, it can be used to save an important shot that simply didn&t stay in the frame one hundred per cent. Realistically, however, we will probably not find such a tool in post-production for the time being, but rather in an Android smartphone - but what is not, can come quickly.

If you&re still not quite sure how VidPanos actually works or what it does, here&s an informative YouTube video from the makers themselves:

more infos at bei vidpanos.github.io

deutsche Version dieser Seite: VidPanos - Google-KI stabilisiert Videos und erfindet fehlende Randbereiche hinzu