In the blink of an eye, the Terracotta Warriors and Horses blinked in the blink of an eye! A foreign guy used AI to restore him

In a thousand years, the moment they blinked, it was as if history was repeating itself.

Not long ago, foreign blogger Panagiotis Constantinou used AI to restore the Qin Dynasty terracotta warriors and horses.

What’s even more amazing is that he also made the ancients make subtle dynamic expressions.

Look, the terracotta warriors and horses are winking and smiling at you.

After visiting the Terracotta Warriors and Horses, everyone will be shocked by these real-life clay sculpture monuments. Look at the Terracotta Warriors and Horses restored with AI, and they turned into flesh-and-blood uncles.

The expressions of each terracotta warriors and horses are all kinds of strange. After thousands of years, it is difficult for us to imagine The way they looked then.

This foreign guy has brought these terracotta warriors back to life and talked to you through thousands of years.

Previously, Otani, the owner of Station B UP, also repaired the terracotta warriors and horses, making their mouths move.

##In fact, in addition to the terracotta warriors and horses, Panagiotis Constantinou also made paintings of Egyptian emperors behind enemy lines, ancient Greek philosophers, the Queen of England, etc. A series of dynamic fixes.

Let’s enjoy it together.

Meet through time and space

Marcus Crassus, the famous richest man in the late ancient Rome

​

Alexander the Great

## Ancient Cleopatra

Queen of England

......

With this series of repairs, foreign bloggers also became popular and became popular. Attracted the attention of many people.

Repair technology

So how is AI repair work generally implemented?

Most of these three technologies are used: the first is DAIN, which is responsible for frame filling, the second is ESRGAN, which is responsible for resolution amplification, and the third is DeOldify, which is used for coloring.

DAIN: Depth-aware video frame interpolation (DAIN) model explicitly detects occlusions by exploring depth information. This project developed a depth-aware flowing projection layer that preferentially samples closer objects to synthesize mid-stream interpolated video frames.

ESRGAN: ESRGAN is improved from SRGAN and is mainly used for video super-resolution.

Contrary to the fact that the deep models of SRGAN are increasingly difficult to train, the deeper ESRGAN models can achieve excellent performance through simple training. The core point lies in the network interpolation strategy that balances visual quality and peak signal-to-noise ratio.

DeOldify: DeOldify is trained using NoGAN, which is crucial for obtaining stable and colorful images.

NoGAN training combines the benefits of GANs (beautiful colorization) while eliminating side effects (like flickering objects in videos).

Whether it is an image or a video, AI can restore it to life, mainly due to its powerful learning ability.

Here we take a brief look at the "soul GAN" that is indispensable for AI repair.

Remember the images of horses and zebras transforming into each other? This is one of the applications of GAN.

Compared with other classic machine learning algorithms, the biggest innovation of the GAN algorithm is that it combines a generator and the discriminator (Discriminator), using the two to compete with each other to achieve mutual optimization, and finally reach a "evenly matched" state.

Take images as an example. After repeated training, the discriminator D hopes to accurately distinguish which is the real data distribution and which is the generator's data distribution. At the same time, the generator G is also continuously trained. , and finally hope to be able to confuse the fake with the real, so that the discriminator cannot determine which image is real and which is fake.

Nowadays, more and more technologies such as AI repair, AI face-changing, etc. are inseparable from the GAN network. The problems that once plagued GAN, such as difficulty in convergence, have gradually been improved, and based on GAN, More and more excellent networks have been derived, such as CycleGAN, StackGAN, etc., and more realistic and vivid images are presented to the public one by one.

Now is probably the era when “everything can be AI”! We will wait and see what kind of visual surprises AI will continue to bring us~~~​

The above is the detailed content of In the blink of an eye, the Terracotta Warriors and Horses blinked in the blink of an eye! A foreign guy used AI to restore him. For more information, please follow other related articles on the PHP Chinese website!