Re-animated History

Source: Deep Learning on Medium

Deep Learning Models in Use

Historical Image, being #DeOldified and #3DKenBurnsed

We have learned a variety of fantastic models, powered by Deep Learning. In perspective, you can call it a data-driven Visualisation Renaissance. But what can we undertake with all these manifold ways for AI image manipulation? Especially if we combine various DL methods.

You can do a time machine.

Here is an interesting example, how quickly one image can become alive in inspiring teamwork. And how unexpected things can happen.

This is Mulberry Street, Manhattan, New York City, once a heart of Little Italy:

Source: (Wiki)

Jason Antic, the developer of DeOldify (DL driven model, which allows colorizing B&W photos), often presents in his twitter timeline the newest results of his work: historical images are getting back their colors and freshness.

Yesterday he posted the image above, being #DeOldified:

Mind the resolution of details and richness of the palette here! You can endlessly contemplate this revival of old Manhatten Street (often referred in movies, books, and films):

Jason, for example, found funny small details (human factor of history is fascinating in every form):

Another twitter user discovered time-traveling Orlando Bloom:

Here you can encounter another glimpse of somebody’s vivid childhood:

Even this: Time machine.

The whole composition almost pleads to be virtualized.

Motioned picture

So I appended the 3D Ken Burns method, developed by Simon Niklaus (read about this method and try it out here):

Or, in better quality:

3D Ken Burns method works at best with such clear vanishing point images. The picture is separated into various spatial layers, and the background voids behind layers are filled up with content-sensitive inpainting.

Such a simulated tracking shot above is convincing. Yet, it can still be improved by more dynamism. Jonathan Fly did it. He fine-tuned the 3D Ken Burns camera settings:

Experimenting with parameters creates new visions:

Yet he even perfectly exaggerated the settings till an exorbitant quality:

You cannot recognize the initial image anymore, but you discover new frontiers behind the image, being dreamed up by Artificial Intelligence:

Thus, in a combination of people’s sharp minds and Deep Learning, the interactive examination of historical footage has become a fascinating exploration of reality limits.

Deep Learning can do way more than opening new frontiers — it can open our minds.