Academia.eduAcademia.edu

Figure from:Parametric Skeletons with Reduced Soft‐Tissue Deformations

See full PDFdownloadDownload figure
Figure 5: This figure shows the distribution of subspace handle points defined for the body (right) and the hand (left) respectively.

Figure 5 This figure shows the distribution of subspace handle points defined for the body (right) and the hand (left) respectively.

Related Figures (16)

Figure 1: We augment parametric skeletal models with reduced soft-tissue deformations. We formulate dynamics in an unposed reference space (left) and design a sparse biharmonic subspace to represent soft-tissue deformation. The local basis efficiently captures inertial effects and external interactions (middle) while preserving accurate data-driven static deformations depending on the skeletal pose. We demonstrate our method with realistic real-time simulations of human characters and hands (right).
Figure 1: We augment parametric skeletal models with reduced soft-tissue deformations. We formulate dynamics in an unposed reference space (left) and design a sparse biharmonic subspace to represent soft-tissue deformation. The local basis efficiently captures inertial effects and external interactions (middle) while preserving accurate data-driven static deformations depending on the skeletal pose. We demonstrate our method with realistic real-time simulations of human characters and hands (right).
Figure 2: This figure highlights the deformations produced by the data-driven skeletal model in our formulation (left: without pose- based blend shapes; right: with pose-based blend-shapes). Our de- formation model matches exactly the data-driven result on static, contact-free poses. J. Tapia, C. Romero, J. Pérez, M. Otaduy / Parametric Skeletons with Reduced Soft-Tissue Deformations
Figure 2: This figure highlights the deformations produced by the data-driven skeletal model in our formulation (left: without pose- based blend shapes; right: with pose-based blend-shapes). Our de- formation model matches exactly the data-driven result on static, contact-free poses. J. Tapia, C. Romero, J. Pérez, M. Otaduy / Parametric Skeletons with Reduced Soft-Tissue Deformations
Figure 3: On a skeletal model such as the one described in Sec- tion 3.1, contact forces are fully absorbed by the skeleton (left). With our augmented soft model described in Section 3.3, contact induces soft-tissue deformations (right).
Figure 3: On a skeletal model such as the one described in Sec- tion 3.1, contact forces are fully absorbed by the skeleton (left). With our augmented soft model described in Section 3.3, contact induces soft-tissue deformations (right).
Figure 4: Volumetric discretization of the soft-tissue layer for both the SMPL (left) and MANO (right) models. Meshes have been gen- erated using a Delaunay tetrahedralization conformal to boundary surfaces [Sil5]. In the former case, we construct the inner sur- face displacing the outer surface with smoothly varying thickness throughout the body. In the latter case, we use the mesh of a real hand skeleton as the inner surface. Our method only needs these meshes during preprocessing, for the computation of the subspace basis and the cubature weights.
Figure 4: Volumetric discretization of the soft-tissue layer for both the SMPL (left) and MANO (right) models. Meshes have been gen- erated using a Delaunay tetrahedralization conformal to boundary surfaces [Sil5]. In the former case, we construct the inner sur- face displacing the outer surface with smoothly varying thickness throughout the body. In the latter case, we use the mesh of a real hand skeleton as the inner surface. Our method only needs these meshes during preprocessing, for the computation of the subspace basis and the cubature weights.
Figure 6: Our soft skeleton body model matches static data-driven deformations accurately for a given pose (left). When in contact with an external object it produces a dynamic response with two- way coupling (right). Soft-tissue deformation is clearly noticeable in the unposed reference shape (inset). Local static deformations due to the pose blend shape (e.g., arm bulging) apply independently of external interactions.
Figure 6: Our soft skeleton body model matches static data-driven deformations accurately for a given pose (left). When in contact with an external object it produces a dynamic response with two- way coupling (right). Soft-tissue deformation is clearly noticeable in the unposed reference shape (inset). Local static deformations due to the pose blend shape (e.g., arm bulging) apply independently of external interactions.
Figure 7: Our handling of inertial forces correctly captures soft- tissue deformations induced by the skeletal motion. The picture shows two frames of an animation with a character subject to ro- tational acceleration (left) and deceleration (right) in the vertical axis, and the resulting colormap of soft-tissue deformations.
Figure 7: Our handling of inertial forces correctly captures soft- tissue deformations induced by the skeletal motion. The picture shows two frames of an animation with a character subject to ro- tational acceleration (left) and deceleration (right) in the vertical axis, and the resulting colormap of soft-tissue deformations.
J. Tapia, C. Romero, J. Pérez, M. Otaduy / Parametric Skeletons with Reduced Soft-Tissue Deformations Figure 8: A parametric approximation of the hand is realistically deformed due to skeletal poses and contact with objects. The pic- ture shows four frames of an animation sequence and a colormap with the corresponding vertex displacements in unposed space (in- sets). Sliding over a structure of fine cylinders produces ripples on the palm (see supplementary video).
J. Tapia, C. Romero, J. Pérez, M. Otaduy / Parametric Skeletons with Reduced Soft-Tissue Deformations Figure 8: A parametric approximation of the hand is realistically deformed due to skeletal poses and contact with objects. The pic- ture shows four frames of an animation sequence and a colormap with the corresponding vertex displacements in unposed space (in- sets). Sliding over a structure of fine cylinders produces ripples on the palm (see supplementary video).
Figure 9: Our model incorporates skeletal tracking of target input animations through control forces. Note that even in the absence of external interactions, dynamic deformations appear as an effect of inertial components. The figure shows a sequence of three animation frames and a colormap with the corresponding vertex displacements in unposed space (insets).
Figure 9: Our model incorporates skeletal tracking of target input animations through control forces. Note that even in the absence of external interactions, dynamic deformations appear as an effect of inertial components. The figure shows a sequence of three animation frames and a colormap with the corresponding vertex displacements in unposed space (insets).
Figure 10: Fingers sliding sideways along a cylindrical shape. The resulting deformation of the fingertips is shown in unposed refer- ence shape. The colormap indicates the first right Cauchy-Green tensor invariant; note the difference between stretch (red) and com- pression (blue) at the sides of the fingertips depending on the effect of sliding direction.
Figure 10: Fingers sliding sideways along a cylindrical shape. The resulting deformation of the fingertips is shown in unposed refer- ence shape. The colormap indicates the first right Cauchy-Green tensor invariant; note the difference between stretch (red) and com- pression (blue) at the sides of the fingertips depending on the effect of sliding direction.
Figure 11: Interactive hand tracking demo using the Leap Motion. The user is capable of touching and grasping environment objects, producing significant soft-tissue deformations as an effect of fric- tional contact.
Figure 11: Interactive hand tracking demo using the Leap Motion. The user is capable of touching and grasping environment objects, producing significant soft-tissue deformations as an effect of fric- tional contact.
Table 1: Performance comparison of a full-space simulation vs. our method, with different numbers of cubature points (denotes as “subspace X”). The table shows average timings per time step, as well as broken down into the evaluation of forces, the evaluation of the energy Hessian, and the linear system solve (all in ms). As shown in Figure 6, the resulting skeletal model accurately matches data-driven static deformations while at the same time re- alistically responding to external interactions. The deformation of the skeleton can be driven by an input animation through tracking forces (Figure 9), and the soft-tissue dynamically deforms accord- ingly due to the effect of inertial forces (Figure 7). Interestingly, as shown in Figure 12, the parametric model allows us to reproduce
Table 1: Performance comparison of a full-space simulation vs. our method, with different numbers of cubature points (denotes as “subspace X”). The table shows average timings per time step, as well as broken down into the evaluation of forces, the evaluation of the energy Hessian, and the linear system solve (all in ms). As shown in Figure 6, the resulting skeletal model accurately matches data-driven static deformations while at the same time re- alistically responding to external interactions. The deformation of the skeleton can be driven by an input animation through tracking forces (Figure 9), and the soft-tissue dynamically deforms accord- ingly due to the effect of inertial forces (Figure 7). Interestingly, as shown in Figure 12, the parametric model allows us to reproduce
Figure 12: Each row presents a sequence of pictures showing two rag-doll characters with different shape parameters undergo the same simulation scenario. The bodies are slowly pushed by a cylinder while hanging from the wrists, resulting in different deformation behaviors.
Figure 12: Each row presents a sequence of pictures showing two rag-doll characters with different shape parameters undergo the same simulation scenario. The bodies are slowly pushed by a cylinder while hanging from the wrists, resulting in different deformation behaviors.
Figure 13: We compare the behavior of our method (top) against the ground-truth full-space problem (bottom). While the full-space solution is slightly smoother, our approximation produces a plausi- ble behavior at a computational cost orders of magnitude lower. whose deformation is controlled by s = 10 shape parameters and p = 144 pose parameters. To define the soft-tissue layer, we follow the approach described in [ROCP20]; here, the inner surface is defined by displacing the vertices of the template mesh a distance
Figure 13: We compare the behavior of our method (top) against the ground-truth full-space problem (bottom). While the full-space solution is slightly smoother, our approximation produces a plausi- ble behavior at a computational cost orders of magnitude lower. whose deformation is controlled by s = 10 shape parameters and p = 144 pose parameters. To define the soft-tissue layer, we follow the approach described in [ROCP20]; here, the inner surface is defined by displacing the vertices of the template mesh a distance
Figure 15: We compare our handle basis (right) and a modal basis (left) on the augmented SMPL model, as a sphere presses the chest of the avatar. Our model resolves the contact smoothly, while the modal basis suffers artifacts both close to the contact and far, in the arm and the elbow. Table 5.2 shows a breakdown of timings, comparing a full- space simulation, our subspace method with the number of cuba- ture points used in the examples, and our method with more cuba- ture points. The subspace formulation provides a speed-up of over 1000 on the linear system solve of each time step. On the other
Figure 15: We compare our handle basis (right) and a modal basis (left) on the augmented SMPL model, as a sphere presses the chest of the avatar. Our model resolves the contact smoothly, while the modal basis suffers artifacts both close to the contact and far, in the arm and the elbow. Table 5.2 shows a breakdown of timings, comparing a full- space simulation, our subspace method with the number of cuba- ture points used in the examples, and our method with more cuba- ture points. The subspace formulation provides a speed-up of over 1000 on the linear system solve of each time step. On the other
Figure 14: We compare our handle basis (center) and a modal basis (right) on the augmented MANO model. As reference, we also show the same scene without soft-tissue deformation (left). The modal basis has difficulty producing the correct bulging on the side of the hand, and i. suffers artifacts on the finger tips, far from the contact area, due to the global support of the basis.
Figure 14: We compare our handle basis (center) and a modal basis (right) on the augmented MANO model. As reference, we also show the same scene without soft-tissue deformation (left). The modal basis has difficulty producing the correct bulging on the side of the hand, and i. suffers artifacts on the finger tips, far from the contact area, due to the global support of the basis.
Figure 16: The density of handles determines the ability to resolve contacts with high spatial frequency. In the images, we press a fin- ger pad of the augmented MANO model with a pin, and we show the handles in black. The contact is resolved more accurately as the pin approaches a handle (left), and smoothly degrades as the pin moves away from the handles (right).
Figure 16: The density of handles determines the ability to resolve contacts with high spatial frequency. In the images, we press a fin- ger pad of the augmented MANO model with a pin, and we show the handles in black. The contact is resolved more accurately as the pin approaches a handle (left), and smoothly degrades as the pin moves away from the handles (right).