PPAP BOXをつくる / モノを認識すること

English follows Japanese



Image-to-Image Demo - Affine Layer








Making PPAP BOX / Recognizing 

Google's artificial neural network learned to detect cats. It is a theme that became a topic in 2012.


In 2017, it became possible to convert a picture of a cat appropriately written into a live-action version cat. The neural network which learned a lot of cat images seems to be able to see anything as a cat. Even if artificial intelligence has only a hammer, everything may look like a nail.

Image-to-Image Demo - Affine Layer

But are these neural networks recognizing "cats"? No, it is not, it only recognizes "cat images". (That's awesome thing to say!) That learned cats from only the outside. But real cats do not sit still, they are crowing, they are heavy if you lift them up, they touch the hair when you touch them, there are smells, and feeling warm when you touch your hands. An image cat is part of a real cat.

Although it is a children's picture book, there is a work called "It Might be an Apple" that Yoshitoshinsuke drew. It is a very pleasant story to imagine that "apples" in the kitchen may actually be "something that looks like apples". But this may be the essence of recognizing things as well.

It is in the kitchen, it is the shape of an apple, it is the color of an apple, it is the weight of an apple, with the same feel as an apple, even if I strike or shake it, it don't make a strange sound, it's not hot or it's not cold, as you approach the nose it smells a little faintly, when you taste a bite and it tastes of apple, it may be an apple

It Might be an Apple

It Might be an Apple


 What I'm going to make is a machine that know apple and pineapple and pen, but surely any way to recognize things with five senses will be able to be recognized by anything. I think that it would be nice to do so.