Contact Us




Contactus:Jon Qiu


Contactus:James Mo



Add:B Building, Laifu Industrial Park, Liyuanxilu, Heping, Fuyong, Baoan, Shenzhen, China.

Home > Knowledge > Content
Do you want to know how you lost Quick, Draw!? Jul 25, 2018

Ladies and gentlemen, these days, the Google AI applet "Quick, Draw!", which has been playing in the circle of friends, have you played? I don't know how the results are. The small card only hangs when it hits Level 3 and loses to a bucket. Maybe you will think how hard it is to draw a bucket? It is really difficult. A seemingly simple game is not at all simple. For example, how does AI guess when it is a banana? Today, we might as well take a look at the hidden mystery behind this little game.

According to Google, "Quick, Draw!" is powered by Google's artificial intelligence neural network technology, which originated from the world's largest data set containing more than 50 million hand-drawn sketches. After machine learning hand-drawn sketches, Objects can be identified by rough sketches, which is why players outline simple lines and AI can quickly guess what you are painting.

In fact, before the release of "Quick, Draw!", Google Brain was launched on a web page called "Quick, Draw!". I guess, based on a build called "sketch-rnn". RNN, the game also randomly displays a noun, asking the player to draw it in 20 seconds, then Quick, Draw! will judge whether you are painting like it. So, what is Quick, Draw!

In fact, both of these games use neural network technology, which is the basis of deep learning and was originally inspired by the biological nervous system to simulate the biological nervous system. In simple terms, AI recognizes the world through neural networks, and is inseparable from CNN convolutional neural networks or RNN cyclic neural networks. The former has been successfully applied to various applications of image recognition, while the latter can use its internal memory to process input sequences of arbitrary timing, so that it is easier to handle handwriting recognition and speech recognition without segmentation.

With the support of neural network technology, and through a large number of graffiti samples in the later period, Quick, Draw! can summarize the feature points of different things. Of course, in addition to these feature points, data training can also give AI a dynamic drawing process, so that AI understands that when humans draw a certain thing, they usually choose what to draw first, how to go, and when to stop. That's why, sometimes when the player just draws the first stroke, the "small song" can give the correct answer. Therefore, the larger the database size used for training, the higher the accuracy of the results given by AI.

That being the case, then why do we sometimes fill the mobile phone screen with AI or can't guess the correct answer? The reason is whether you have drawn the core features of things.

The key to promotion is to draw "core features"

Like the pear below, when we draw a sketch, the AI gives the nose, cello, violin multiple options under the sketch, but there is no pear. . . Explain that in the sketch drawn earlier, there is no core feature that allows the AI to recognize the "pear". Therefore, I would like to think more about what others will draw when they draw pears. How to draw can make AI think that you are painting pears, not violins and noses!

Quick, Draw!1.jpg

Or, when we draw a fish, it may look like this, and every detail is very detailed. ↓

Quick, Draw!2.jpg

However, after Quick, Draw! observed a lot of players' handwriting, the little song thought that the fish should be like this. . ↓

Quick, Draw!3.png

Obviously, in the case of Quick, Draw!, the player's fish painting is too complicated. Therefore, it is not difficult to find that in addition to quickly drawing the core features of the object, it is best to outline a rough shape (or outline). After all, the number of alternative answers is limited. .

     There are still many people who haven't painted it because the data cable is too inconvenient when charging the phone. So, isn't it too fast to use the WST wireless charger? We offer wireless charging PAD: WST-DIX and wireless power bank: WST-AirDuo.