An active area of research is to use generative network to model videos so that simple linear interpolation of the hidden state can lead to meaningful transition between frames.
ljelenak
Can you sometimes just replace key frames like this one with physical simulation?
keenan
@anonymous_panda Interesting. Though also: dangerous! In the future, how will we know whether we're watching events that really happened, or interpolated motion that was dreamed up by a deep neural net? :-)
keenan
@ljelenak Yes, absolutely. A bouncing ball, for instance, obviously makes sense to animate via physics. Though if you want really "cartoony" animation (e.g., the ball squashes in an exaggerated way when it lands, or stretches when it springs into the air) you may still want to animate by hand. This is in fact one of the major techniques in hand-drawn animation: squash and stretch
An active area of research is to use generative network to model videos so that simple linear interpolation of the hidden state can lead to meaningful transition between frames.
Can you sometimes just replace key frames like this one with physical simulation?
@anonymous_panda Interesting. Though also: dangerous! In the future, how will we know whether we're watching events that really happened, or interpolated motion that was dreamed up by a deep neural net? :-)
@ljelenak Yes, absolutely. A bouncing ball, for instance, obviously makes sense to animate via physics. Though if you want really "cartoony" animation (e.g., the ball squashes in an exaggerated way when it lands, or stretches when it springs into the air) you may still want to animate by hand. This is in fact one of the major techniques in hand-drawn animation: squash and stretch
More fun examples here: https://blog.animationmentor.com/squash-and-stretch-the-12-basic-principles-of-animation/