A movie as if Artificial Intelligence uses LSD to see "Bob's painting classroom"

Artificial intelligence created by Google "DeepDream"Finds objects that are familiar even a little from the image and reconstructs and outputs it, but the image to be created became a topic of" a nightmare seems to be a subject "for a while. Alexander Lieben, an engineer, uses AI like DeepDreamBob RothMovie rebuilt "Bob's painting classroom"Deeply Artificial Trees"Has been released,LSDIt has become a terrible movie as if you are watching "Bob's painting classroom" after using.

Deeply Artificial Trees on Vimeo


A terrible painting is projected ... ...


A humanoid silhouette appears before the canvas.


Fill the canvas ...


The original title of Bob's painting classroom "The Joy of Painting" is displayed.


And Bob Ross appears ... ... but AI is transforming that figure into a terrible one. Also, because speech is arranged by AI, what you are talking about can not be heard well. However, it seems that the talking about "English-like something" is transmitted.


I start to extend the paint on the palette with my brush, but at a level where I do not know where the hand is and where it is a brush if it is a still picture.


When the upper body is reflected, you can see that it is a human type at first.


The scene changes when I start drawing something ......


I am touching something on the palm of your hand.


It will be a hand up, but it is unknown what is touching.




Furthermore, it interacts with a quadruped walking animal like a deer ......


I returned to the painting scene.


It is as if expressing a nightmare on a canvas.



I understand that human beings are drawing pictures when drawing pictures ......


What on earth were you drawing? ...?


EngadgetAccording to Mr. Leben's audio output of AI createdWaveNetHe was in charge of that, he gave voice data of "Bob's painting classroom" and let him learn. According to Mr. Leven, "I am basically designing to create a model that has captured audio and heard and created new audio based on that model." The image output part of AI created by Mr. Leben seems to use two sets of machine learning algorithms "DeepDream and TensorFlow" and "Keras and VGG".

Interestingly, AI responsible for audio output and video output is produced separately. In order to finish the audio output to a certain level, it seems that it was necessary to prepare the audio data of "Bob's painting classroom" for full season in AI's learning, but it is necessary to prepare for AI learning which is responsible for video output only a few It seems that only episodes were needed, and the difference was also very interesting.