Our next topic is procedural deformation. So, the work we introduce here is called ProcDef or Local-to-global Deformation for Skeleton-free Character Animation. So the goal here is to design a flexible character animation such as jelly fish or worm, or intestine, like a heart. So important point here is that they are full body motion. You know, entire body, every detail is moving, and also there is a dynamic. You know, it, it's not, how to say freeze dry animation. It reacts dynamic. If you push, then the formation changes, and so on. So there are traditional methods. These are the traditional methods. One is skeleton-based. This is most popular skeleton-based control, like a joint angle control, just we showed in the last video. This is, technique is useful for character animation like a human or animals. But you know, if there is no obvious skeleton structure, like a jellyfish, this technique doesn't work. Another possible approach is keyframe-based. So, for each timeframe, you specify the shape individually and the system just interpolate them. However, in order to do this, you have to specify many, many key poses manually by pushing around many vertices, which is very tedious. And more importantly, this kind of keyframe-based motion cannot calc- handle dynamics, you know. The motion is kind of fixed animation and it cannot respond to the forces from the environment. So our approach is to procedural approach. So the user specifies how local regions of a entire body stretch or shrink. And then by assembling them together system generates a global motion. And the user also designs the patterns of the local deformations. Like, this is time sequence, you know, stretch, shrink, stretch, shrink, very quickly. Or slowly, or fast stretch and slowly shrink, something like that. So let me show you a video. [BLANK_AUDIO] Hm, yes. [BLANK_AUDIO] So here is our result. So this kind of animation is our goal, the full body animation of a flexible body characters. So the first task is design orientation field. So given a 3D shape, user first specifies an orientation field. [BLANK_AUDIO] So here the user specifies the layer structure here actually, so the red is inside and the blue is outside. So in this way, user specifies the layer structure. And you will see it like this. So after user paints red inside, blue outside, system interpolates them to compute a continuous layers. And after specifying layers, user specifies orientation of vector field on the layer surface. So in here you can specify, the user specifies rotational flow field. So this is a flow field. So now, you have a flow field along this layer surface, and then layer surface directional vectors, gradient vector. And then you can get the cross-border app to get a free frame field. So, in this way, you get a global representation field, inside of an object. Next we define how each individual element deforms depending on the time. [BLANK_AUDIO] So, in this example we have three fields, this primary direction and second direction, and third direction. And for each direction, we specify the pattern, motion pattern. So the top one specifies a change in the Lengths, direction and then if you longer and shorter, longer and shorter. [BLANK_AUDIO] So, yeah, in this way the user specifies a change in the local shape and then global shape is obtained as a result of the smaller motions. [BLANK_AUDIO] You can easily represent this kind of squashing operations using this one So, the previous example, all elements move simultaneously in the same way. But in order to make more interesting eh, motion, you can specify the amplitude field. So here red and blue has a opposite direction opposite amplitude. So in this way, if the blue stretches, red shrinks, and red stretches, blue shrinks. So in this way, you can get this kind of bending motion. And the previous one is a fixed body for the entire body. But the next one is propagation. So in this way you can introduce delay. So the formation starts from here and then slowly later slightly later, the formation propagates here and later and later. So there's a delay in the propagation and then you will get this kind of, of motion. Propagation of motion. So by combining this kind of phase shift and amplitude, and orientation field you can get leech infor- leech animations. So here is a couple of results. So this is a worm walking around. And the important thing here is that this is dynamically computing. Dynamically computed considering physics, so as the user interacts with it, the system continues animation with deformation, but also responding to the environment. This is not animational pre-programmed fixed animation. This is different way of making a motion. [BLANK_AUDIO] So, there's another sample as you see we have a collision detection continuously running. So they're individually moving around, but also collide with each other and then animation changes, depending on the collision. And this demonstration is to demonstrate how fast the algorithm is. [BLANK_AUDIO] And here's another example of a response to user input. So as the user touches, and then deformation propagates starting from the user interaction. So you can get this kind of interesting motion. Yeah, this is a propagation. This an example of heart animation. The user manipulates it. But this is orientation field. [BLANK_AUDIO] And then you can get the heart works as a pump. [BLANK_AUDIO] So that's it. So as I said, deformation design consists of two phases. One is the design orientation field. You first specify a layer structure and then specify an orientation field. And then after specifies orientation, the user specifies for each orie- each direction. User specifies a stretch and shrink pattern, time sequence pattern. So now we will, I will describe the algorithm to synthesize global motion from a speci- user-specified local motions. And in the preprocessing so the algorithm consists of preprocessing part and animation synthesis process. So in the preprocessing, the user inputs, input is a tetrahedral volumetric mesh. And having tetrahedral mesh for individual vertex, individual vertex and we compute a local region, N. Like this one for individual vertex by connecting a 1-ring. 1-ring is a connection of vertices next and neighboring to each individual vertex. And after having this local regions in the animation process, we deform local region N depend- depending on the orientation field and also the timing chart, you know the orientation field defines the orientation of the stretch squash and then times sequence chart defines, time sequence pattern defines whether to stretch or shrink. So in this way, you first deform the individual regions, one by one, stretch or shrink. And after that, the system assemble them together to get a global shape. So that's the basic form. Let me describe a little bit more detail. So this is original rest shape. And then this is the current deformed shape in physical simulation. And then the task is to define the motion of the next time frame, starting from this current time frame shape. So let me describe how to do it. So in the original shape, you first compute the local regions in the free computation. And for each local region, we compute the next time step local regional target shape. Some part should be shrinked, some parts will be squa- stretched, and so. So this is a deformed local shape. And after that, what the system do is shape matching. This is the most important part. Shape matching. For each local region, we try to find the most, best matching position. You know, if you get these three, the system tries to match these three by rotating like this way. So, shape matching tries to find the best matching positional change, rotation and translation. And after matching, you know the current position and desired position. So the system applies force from the current position to the matched desired position. Defined by the deformed local shape. And then apply force. After that this is just the standard physical simulation. Applied force move towards it, applied force moves towards the position. So here's a summary. So I just introduced animation of deformable characters like jellyfish, a worm, and so on. And the algorithm is a local to global approach, so user specifies orientation field and then deformation pattern, and then each local region changes accordingly, and then global motion appears as a result of aggregation. And inside we use a shape matching dynamics, because it's very fast and stable, fast and stable. So to learn more, original paper was published as procedural deformation ProcDef, Local-to-global Deformation for Skeleton-free Character Animation. And design of orientation is actually discussed in a separate paper called Lapped Solid Textures: Filling a Model with Anisotropic Textures. So, they used an orientation field to put, paste textures inside of an object. And that, dynamic simulation we use is called shape matching. And this is very popular technique in graphic community, because it's, it's not necessarily physically accurate, but it's very, very stable. Depending even if the user input is very eh, contradictory, and so on. And I recommend you to take a look at these papers, because they can be very useful. Thank you.
