Tokyo Gentleman Story

Chapter 1696: How can this be called a dirty deal!

, The fastest update to the latest chapter of Tokyo Gentleman Story!

"It's not just a virtual idol. I think if Mori-kun thinks it is good, it would be great for movies."

Ayana didn't notice Morinatsu's thoughtfulness, but continued to talk.

This technology in film and television can be used to denoise, and the accuracy requirements of the film itself are different from those in military or precision technology, so the algorithm can be much looser, and the accuracy does not need to be considered. As long as the picture is beautiful, it is enough Up.

By the way, Disney has made great achievements in this area, and on the other world line, the facial expression capture of Avatar is applied to related technologies.

As for Senxia, what I want is relatively simple, let's take the problem to be solved here.

Hmm...just cut out the picture.

In fact, the so-called motion capture, in a narrow sense, is actually a cutout.

Convolutional neural network matting works well. It's like lifting weights lightly. In the hands of trained artificial intelligence, it is really very good.

But Ayana’s technique, if only used to cut out pictures...it seems to be a bit overkill...

"Ayana, you really didn't think about being a programmer... No, I mean, did you really think about doing computer research?"

Senxia finally asked.

However, the latter firmly raised his head: "I want to be a voice actor."

Ok.

Ayana's voice is actually quite distinctive.

"Ah... but you are really promising..."

"I, want, be, voice actress!"

Ayana paused.

Senxia closed her mouth obediently.

Well, this issue should not be debated at this time.

Moreover, Ayana is willing to contribute to the career of a gentleman, so she should be happy...

But it's a pity, if Ayana is willing to do programs and write algorithms, that would be even better.

"Where are we talking?" Senxia decided to bring the topic back first.

"The application of film and television." Ayana said.

"Oh, yes, do you think this can also be used?" Senxia asked.

"Of course, your movie will be shown next year."

Outsiders know less about the big movies in Hollywood, but the people around Morika still know a little.

In fact, in many cases, the purpose of neural networks is not to replace real people for work, but to reduce people's repetitive work.

As for the neural network itself, it is a "robot" that requires deep learning.

The "large amount of training" mentioned by Ayana just now refers to the learning data here. The input of the data is a dirty point cloud calculated by the computer, and the output is a regular grid calculated by the computer after manual processing-this is all from the results of computer calculations, not from actual data .

In this way, AI can gradually understand how people work, and then they can also "imitation" themselves.

This feeling, if you insist on finding a prototype, is equivalent to learning some similar life habits of the owner under the influence of your own kittens and puppies, and then these kittens and puppies will act like their owners. It's deep learning.

Of course, the convolutional neural network here cannot essentially reduce the production cost, because both the training data and the final application come from a very expensive system. To put it another way, if there is no such expensive system and only this neuron network, it basically means nothing.

In another world line, there is a software for coloring and changing faces. This type of software requires the user to train the AI with thousands of photos before the machine starts to work, and finally let the machine know what color and face to change for what composition.

There used to be a dear friend who took many photos of his crush, and then trained his AI, and then asked the AI to replace the actresses in love action movies with the other's appearance.

then……

Then this guy began to doubt life.

This kind of person who actively puts a green hat on his head is really rare in the world.

"If the cost of the terminal can be reduced, maybe it's not bad..."

For Senxia, it does not matter if the overall cost is not reduced.

What Morinia needs here is the mere R&D cost. What he cares more about is the cost of the terminal.

If this method can greatly reduce the cost of players and software developers themselves, it is naturally more worthwhile.

What's more, the cost of a trained model is fixed. In other words, the more you sell, the lower the cost.

Ayana also knows this, so he came up with this neural network.

Of course, the key is this "training" process.

The most difficult part of matting is to dig out the details. It takes a PS master to dig out an original picture, and it takes the same time to dig out a picture that has been rough-cut by an algorithm. This is because the details are the most complicated place and can be determined. All success or failure. Only trained AI can perform this work in detail.

For example, for example, train an AI with a picture of the moon, and then give a light bulb to the AI, and the AI can optimize the light bulb into the moon...

Well, this is an example of deep learning.

"I can do only limited things in this regard, but I can give you some suggestions and directions. Oh, of course, this is just my personal experience, it's your own business."

"Can't you help me improve?" Senxia asked.

"No time." Ayana shook her head. "It's an idol activity again, and it's an animation dubbing. A holiday like today is actually quite rare."

But it still feels quite a pity...

Eh... not right.

"Ayana." Morinya suddenly thought of something, with a sly smile on her face, "I think this job belongs to you."

"Sorry, but it really doesn't work." Ayana said that she didn't buy in.

"The heroine."

However, Morinatsu's words made Ayana froze.

"At least the female number two, and I can also ask you to dub the heroine."

The big boss spoke!

Hearing these words, Ayana's body began to tremble.

She... shaken!