We present GaVA-CLIP, a knowledge augmentation strategy for Gait Video Analysis, designed to assess diagnostic groups and gait impairment. Based on the large-scale pretrained Vision Language Model, CLIP, GaVA-CLIP learns and enhances visual, textual, and numerical representations of patient gait videos through collective learning across three distinct modalities: gait videos, class-specific descriptions, and numerical gait parameters. Our specific contributions are two-fold: First, we adopt a knowledge-aware prompt tuning strategy to utilize class-specific medical descriptions in guiding text prompt learning. Second, we integrate paired gait parameters as numerical texts to enhance the numeracy of textual representations. Results demonstrate that GaVA-CLIP not only significantly outperforms state-of-the-art (SOTA) methods in video-based classification tasks but also adeptly decodes the learned class-specific text features into natural language descriptions using the vocabulary of quantitative gait parameters. The code and associated clinical knowledge are available at: https://github.com/lisqzqng/GaVA-CLIP.

Objective gait assessment in Parkinson’s Disease (PD) is limited by the absence of large, diverse, and clinically annotated motion datasets. We introduce Care-PD, the largest publicly available archive of 3D mesh gait data for PD, and the first multi-site collection spanning 9 cohorts from 8 clinical centers. All recordings (RGB video or motion capture) are converted into anonymized SMPL meshes via a harmonized preprocessing pipeline. Care-PD supports two key benchmarks: supervised clinical score prediction (estimating Unified Parkinson’s Disease Rating Scale, UPDRS, gait scores) and unsupervised motion pretext tasks (2D-to-3D keypoint lifting and full-body 3D reconstruction). Clinical prediction is evaluated under four generalization protocols: within-dataset, cross-dataset, leave-one-dataset-out, and multi-dataset in-domain adaptation.To assess clinical relevance, we compare state-of-the-art motion encoders with a traditional gait-feature baseline, finding that encoders consistently outperform handcrafted features. Pretraining on Care-PD reduces MPJPE (from 60.8mm to 7.5mm) and boosts PD severity macro-F1 by 17\%, underscoring the value of clinically curated, diverse training data. Care-PD and all benchmark code are released for non-commercial research (Code, Data).

Abstract Human motion generation is a key task in computer graphics. While various conditioning signals such as text, action class, or audio have been used to harness the generation process, most existing methods neglect the case where a specific body is desired to perform the motion. Additionally, they rely on skeleton‐based pose representations, necessitating additional steps to produce renderable meshes of the intended body shape. Given that human motion involves a complex interplay of bones, joints, and muscles, focusing solely on the skeleton during generation neglects the rich information carried by muscles and soft tissues, as well as their influence on movement, ultimately limiting the variability and precision of the generated motions. In this paper, we introduce Shape‐conditioned Motion Diffusion model (SMD), which enables the generation of human motion directly in the form of a mesh sequence, conditioned on both a text prompt and a body mesh. To fully exploit the mesh representation while minimizing resource costs, we employ spectral representation using the graph Laplacian to encode body meshes into the learning process. Unlike retargeting methods, our model does not require source motion data and generates a variety of desired semantic motions that is inherently tailored to the given identity shape. Extensive experimental evaluations show that the SMD model not only maintains the body shape consistently with the conditioning input across motion frames but also achieves competitive performance in text‐to‐motion and action‐to‐motion tasks compared to state‐of‐the‐art methods.

Facial expression generation is one of the most challenging and long-sought aspects of character animation, with many interesting applications. The challenging task, traditionally having relied heavily on digital craftspersons, remains yet to be explored. In this article, we introduce a generative framework for generating 3D facial expression sequences (i.e., 4D faces) that can be conditioned on different inputs to animate an arbitrary 3D face mesh. It is composed of two tasks: (1) learning the generative model that is trained over a set of 3D landmark sequences and (2) generating 3D mesh sequences of an input facial mesh driven by the generated landmark sequences. The generative model is based on a Denoising Diffusion Probabilistic Model (DDPM), which has achieved remarkable success in generative tasks of other domains. While it can be trained unconditionally, its reverse process can still be conditioned by various condition signals. This allows us to efficiently develop several downstream tasks involving various conditional generations, by using expression labels, text, partial sequences, or simply a facial geometry. To obtain the full mesh deformation, we then develop a landmark-guided encoder-decoder to apply the geometrical deformation embedded in landmarks on a given facial mesh. Experiments show that our model has learned to generate realistic, quality expressions solely from the dataset of relatively small size, improving over the state-of-the-art methods. Videos and qualitative comparisons with other methods can be found at https://github.com/ZOUKaifeng/4DFM . Code and models will be made available upon acceptance.

Abstract The capability to generate simulation‐ready garment models from 3D shapes of clothed people will significantly enhance the interpretability of captured geometry of real garments, as well as their faithful reproduction in the digital world. This will have notable impact on fields like shape capture in social VR, and virtual try‐on in the fashion industry. To align with the garment modeling process standardized by the fashion industry and cloth simulation software, it is required to recover 2D patterns, which are then placed around the wearer's body model and seamed prior to the draping simulation. This involves an inverse garment design problem, which is the focus of our work here: Starting with an arbitrary target garment geometry, our system estimates its animatable replica along with its corresponding 2D pattern. Built upon a differentiable cloth simulator, it runs an optimization process that is directed towards minimizing the deviation of the simulated garment shape from the target geometry, while maintaining desirable properties such as left‐to‐right symmetry. Experimental results on various real‐world and synthetic data show that our method outperforms state‐of‐the‐art methods in producing both high‐quality garment models and accurate 2D patterns.

We present a generative model for the 3D facial expression mesh sequences, from onset to the termination of a desired expression. We tailor a Transformer VAE architecture: The encoder compresses a sequence of facial landmarks into an expression-aware regularized latent space, while the decoder generates a new sequence from the sampled latent variable, conditioned on a desired expression. After a landmark-guided mesh deformation, a given 3D neutral face is driven to an animated mesh sequence with the expected expression. The generated sequences are consistent, of quality, and exhibit a good level of diversity, improving over state-of-the-art methods. We validate our model by conducting extensive experiments on two representative datasets. The supplementary video and code are available on a GitHub page (https://github.com/ZOUKaifeng/FacialExpressionGeneration).

Abstract While collections of scan shapes are becoming more prevalent in many real‐world applications, finding accurate and dense correspondences across multiple shapes remains a challenging task. In this work, we introduce a new approach for refining non‐rigid correspondences among a collection of 3D shapes undergoing non‐rigid deformation. Our approach incorporates a Region Of Interest (ROI) into the refinement process, which is specified by the user on one shape within the collection. Based on the functional map framework and more specifically on the notion of cycle‐consistency, our formulation improves the overall matching consistency while prioritizing that of the region of interest. Specifically, the initial pairwise correspondences are refined by first defining the localized harmonics that are confined within the transferred ROI on each shape, and subsequently applying the CCLB (Canonical Consistent Latent Basis) framework both on the global and the localized harmonics. This leads to an enhanced matching accuracy for both the ROIs and the overall shapes across the collection. We evaluate our method on various synthetic and real scan datasets, in comparison with the state‐of‐the‐art techniques.

With the recent advances in data acquisition techniques, the compression of various dynamic mesh sequence data has become an important topic in the computer graphics community. In this article, we present a new spatio-temporal segmentation-based approach for the adaptive compression of the dynamic mesh sequences. Given an input dynamic mesh sequence, we first compute an initial temporal cut to obtain a small subsequence by detecting the temporal boundary of dynamic behavior. Then, we apply a two-stage vertex clustering on the resulting subsequence to classify the vertices into groups with optimal intra-affinities. After that, we design a temporal segmentation step based on the variations of the principal components within each vertex group prior to performing a PCA-based compression. Furthermore, we apply an extra step on the lossless compression of the PCA bases and coefficients to gain more storage saving. Our approach can adaptively determine the temporal and spatial segmentation boundaries to exploit both temporal and spatial redundancies. We have conducted extensive experiments on different types of 3D mesh animations with various segmentation configurations. Our comparative studies show the advantages of our approach for the compression of 3D mesh animations.

With the recent advances of data acquisition techniques, the compression of various 3D mesh animation data has become an important topic in computer graphics community. In this paper, we present a new spatio-temporal segmentation-based approach for the compression of 3D mesh animations. Given an input mesh sequence, we first compute an initial temporal cut to obtain a small subsequence by detecting the temporal boundary of dynamic behavior. Then, we apply a two-stage vertex clustering on the resulting subsequence to classify the vertices into groups with optimal intra-affinities. After that, we design a temporal segmentation step based on the variations of the principle components within each vertex group prior to performing a PCA-based compression. Our approach can adaptively determine the temporal and spatial segmentation boundaries in order to exploit both temporal and spatial redundancies. We have conducted many experiments on different types of 3D mesh animations with various segmentation configurations. Our comparative studies show the competitive performance of our approach for the compression of 3D mesh animations.

Abstract As the symbol of beauty, floral objects have been one of the most popular subjects of artistic drawing. However, designing 3D floral models is generally time‐ and resource‐consuming, because of their structural and geometrical complexity. In this paper, we address the problem of reconstructing floral objects from sketch input. The user draws a relatively clean sketch of a flower and a few additional guide markings from an arbitrary view to rapidly create quality geometric models of flowers. Our system offers a novel modeling scheme compared to several existing flower modelers accepting sketch as input, where the user is required to work with different views, providing step‐by‐step sketch input. Given the silhouette and the guide strokes, an assumed, common botanical structure is estimated, i.e. a cone for each ring of petals. The cones and the silhouette sketch that we segment into elementary curves are used to retrieve model elements from the pre‐constructed shape database. These elements are then placed together around the cone, where an additional, per‐element deformation is performed so as to maximize the silhouette similarity between the user sketch and the 3D flower model from the chosen view. Our system has shown to robustly create a variety of flowers in various configurations, including flower models with several petal layers and various blooming degrees, drawn from different views.

Abstract Understanding the attentional behavior of the human visual system when visualizing a rendered 3D shape is of great importance for many computer graphics applications. Eye tracking remains the only solution to explore this complex cognitive mechanism. Unfortunately, despite the large number of studies dedicated to images and videos, only a few eye tracking experiments have been conducted using 3D shapes. Thus, potential factors that may influence the human gaze in the specific setting of 3D rendering, are still to be understood. In this work, we conduct two eye‐tracking experiments involving 3D shapes, with both static and time‐varying camera positions. We propose a method for mapping eye fixations (i.e., where humans gaze) onto the 3D shapes with the aim to produce a benchmark of 3D meshes with fixation density maps, which is publicly available. First, the collected data is used to study the influence of shape, camera position, material and illumination on visual attention. We find that material and lighting have a significant influence on attention, as well as the camera path in the case of dynamic scenes. Then, we compare the performance of four representative state‐of‐the‐art mesh saliency models in predicting ground‐truth fixations using two different metrics. We show that, even combined with a center‐bias model, the performance of 3D saliency algorithms remains poor at predicting human fixations. To explain their weaknesses, we provide a qualitative analysis of the main factors that attract human attention. We finally provide a comparison of human‐eye fixations and Schelling points and show that their correlation is weak.

The use of cost-efficient and portative devices, such as the Ki-nect, has facilitated the reconstruction of indoor scenes, which has become a field of interest to many. While a large number of applications have been developed for the accurate reconstruction of a scene through efficient registration of multiple scans, the scanning of the scene by the user remains a time-consuming process. This is particularly the case when the scan must not contain any missing data. In this paper, we propose a method requiring only a few shots without any overlapping requirement, which makes the scanning process very simple and light. Given a set point clouds (shots) sparsely taken from an indoor scene, our method seeks for transformations that align the shots together, creating an alignment graph at the same time. It then searches within this graph for the set of transformations corresponding to a possible solution, providing a reconstruction of the scene. Since our method allows missing views and can find all plausible solutions, it can be used to create new scenes by assembling shots obtained from different real or virtual scenes.

This paper presents an image-based rendering (IBR) system based on RGB-D images. The input of our system consists of RGB-D images captured at sparse locations in the scene and can be expanded by adding new RGB-D images. The sparsity of RGB-D images increases the usability of our system as the user need not capture a RGB-D image stream in a single shot, which may require careful planning for a hand-held camera. Our system begins with a single RGB-D image and images are incrementally added one by one. For each newly added image, a batch process is performed to align it with previously added images. The process does not include a global alignment step, such as bundle adjustment, and can be completed quickly by computing only local alignments of RGB-D images. Aligned images are represented as a graph, where each node is an input image and an edge contains relative pose information between nodes. A novel view image is rendered by picking the nearest input as the reference image and then blending the neighboring images based on depth information in real time. Experimental results with indoor scenes using Microsoft Kinect demonstrate that our system can synthesize high quality novel view images from a sparse set of RGB-D images.

We describe a method for modeling helices from planar curves. Given a polygonal curve in the ( x , y ) plane, the method computes a helix such that its orthogonal projection onto the ( x , y ) plane fits the polygonal curve . The helix curve is first sampled and the transformation matrix that best aligns points of the sampled helix to those of the polygonal curve is calculated. This transformation matrix is then used to estimate the parameters of the helix whose projection fits the polygonal curve.

Abstract This paper presents a new technique which makes use of deformation and motion properties between animated meshes for finding their spatial correspondences. Given a pair of animated meshes exhibiting a semantically similar motion, we compute a sparse set of feature points on each mesh and compute spatial correspondences among them so that points with similar motion behavior are put in correspondence. At the core of our technique is our new, dynamic feature descriptor named AnimHOG, which encodes local deformation characteristics. AnimHOG is ob‐tained by computing the gradient of a scalar field inside the spatiotemporal neighborhood of a point of interest, where the scalar values are obtained from the deformation characteristic associated with each vertex and at each frame. The final matching has been formulated as a discreet optimization problem that finds the matching of each feature point on the source mesh so that the descriptor similarity between the corresponding feature pairs as well as compatibility and consistency as measured across the pairs of correspondences are maximized. Consequently, reliable correspondences can be found even among the meshes of very different shape, as long as their motions are similar. We demonstrate the performance of our technique by showing the good quality of matching results we obtained on a number of animated mesh pairs.

We present an efficient and robust algorithm for the landmark transfer on 3D meshes that are approximately isometric. Given one or more custom landmarks placed by the user on a source mesh, our method efficiently computes corresponding landmarks on a family of target meshes. The technique is useful when a user is interested in characterization and reuse of application-specific landmarks on meshes of similar shape (for example, meshes coming from the same class of objects). Consequently, across a set of multiple meshes consistency is assured among landmarks, regardless of landmark geometric distinctiveness. The main advantage of our method over existing approaches is its low computation time. Differently from existing non-rigid registration techniques, our method detects and uses a minimum number of geometric features that are necessary to accurately locate the user-defined landmarks and avoids performing unnecessary full registration. In addition, unlike previous techniques that assume strict consistency with respect to geodesic distances, we adopt histograms of geodesic distance to define feature point coordinates, in order to handle the deviation of isometric deformation. This allows us to accurately locate the landmarks with only a small number of feature points in proximity, from which we build what we call a minimal graph. We demonstrate and evaluate the quality of transfer by our algorithm on a number of Tosca data sets.

ABSTRACT We describe a compression method for three‐dimensional animation sequences that has notable advantages over existing techniques. We first aggregate the frame data by similarity and reorganize them into clusters, which results in the sequence split into several motion fragments of varying lengths. To minimize the number of clusters and obtain optimal clustering, we perform frame alignment, which eliminates the “global” rigid transformation from each frame data and use only “pose” when evaluating the similarity between frames. We then apply principal component analysis for each cluster, from which we get coordinates of corresponding frames in a reduced dimension. Because similar frames are considered, the number of coefficients required for each frame becomes smaller; thus, we obtain better dimension reduction for a given reconstruction error. Further, we perform intracluster compression based on linear coding. Because every motion fragment presents similar frames, conventional linear predictive coding can be replaced by key frame‐based linear coding to achieve minimal reconstruction error. Results show that our method can obtain a high compression ratio, with a limited reconstruction error. Copyright © 2013 John Wiley & Sons, Ltd.

This paper presents a system to create mirror-symmetric surfaces from free-form sketches. The system takes as input a hand-drawn sketch and generates a surface whose silhouette approximately matches the input sketch. The input sketch typically consists of a set of curves connected at their endpoints, forming T-junctions and cusps. Our system is able to identify the skewed-mirror and translational symmetry between the hand-drawn curves and uses this information to reconstruct the occluded parts of the surface and its 3D shape.

Abstract We introduce in this paper a new method for smooth foldover‐free warping of images. It allows users to specify the constraints in two different ways: positional constraints to constrain the position of points in the image and gradient constraints to constrain the orientation and scaling of some parts of the image. We then show how our method is used for texture mapping with hard constraints. We start with an unconstrained planar embedding of the target mesh calculated with conventional methods. In order to obtain a mapping that satisfies the user‐defined constraints, we use our warping method to align the features of the texture image with those of the unconstrained embedding. Compared to previous work, our method generates a smoother texture mapping and offers higher level of control for defining the constraints.

Abstract Intuitive human‐computer interaction has been an active research area with the participation of both academic and industrial groups. It is an important technical and psychological element in computer games, digital product design, cybersports, etc. In this paper, we present our work on a 3D billiards game featuring a haptic interface. Most of the existing games available today concentrate on the development of sophisticated physics engines to obtain realistic motion of billiards balls or of sound engines for realistic sound effect. Our focus in this work is on the improved realism on the hitting action of the game with the aid of haptic device. We do so by computing accurate feedback force on the haptic handle, considering the exact point of contact and physically faithful collision response. Also presented is the physically based modeling of the game environment, including the collision interaction among billiards balls, and between the table and billiards balls. As a result, the players can enjoy a more realistic and pleasurable game compared to other existing 3D billiards games. Copyright © 2010 John Wiley & Sons, Ltd.

This article presents a new approach for Fourier-based modeling of bone anatomies for compression and smoothing. By treating the bone surface as a level set of a 3D scalar field, we can model topologically complex models such as bones. In particular, we experiment with five different alternatives, and prove that 3D scalar field which allows monotonous continuity around the boundary can be a good choice for volumetric description of the surface. This allows avoiding Gibbs phenomena which previous volume-based methods have suffered from, returning better results in compression and smoothing than other scalar fields. We demonstrate the efficacy of the proposed method by showing results with various bone data.

In this work, we record a dancer's rhythmic movement with background music. The captured motion sequences are then segmented into dozens of motion clips, to construct a motion database consisting of sets of labeled motion clips. Many of these motion clips contain short and rapid transition from one main dancing motion to another, which causes unnatural, awkward movements when they are connected in different orders than the original sequence. In this paper, we describe our approach for automatically detecting the transition parts in the segmented motion clips. For each motion clip, we model the motion data using the Gaussian mixture model (GMM) and use the resulting distribution cluster map to improve the efficiency and convergence of the clustering, principal component analysis (PCA) has been applied to the motion data prior to performing GMM. Experiments and comparative analysis show that this PCA-based GMM method effectively performs transition detection on the segmented motion clips.

Tight-fit cloth pressure provides important clue on how well a cloth fits to a body and thus on how comfortable the wearer feels with the cloth. Traditionally-used pressure sensor devices are expensive, sensitive to the experimental environment, and difficult to reproduce. In this paper, a physically-based cloth simulator has been tested for its usability as to measuring the cloth pressure, in order to replace physical measurement of cloth pressure that requires careful operation of pressure sensors. We use existing cloth simulator based on a particle system and measure spring forces exerted on each particle along its normal direction, divided by the summed area of triangles adjacent to that particle. To quantitatively validate the pressure values from the simulator, we have conducted comparative analysis on a set of thin-shell cylindrical tubes --- clothing pressure values have been measured by theoretical estimation and physical experiments using pressure sensors, and compared with those measured by the simulation. While their absolute pressure values differ from each other they exhibit a consistent tendency. From these comparative studies we concluded that cloth simulator can actually be used to measure tight-fit cloth pressure, and further conducted the clothing pressure measure on 3D human body models using the simulator.

Abstract The advent of 3D scanning technology has allowed effective measurement and analysis of breast size and shape, attracting interests by plastic surgeons, brassier designers, etc. Much work remains, however, before 3D scanning systems can be successfully used in automated analysis and synthesis of the breast—filtering noise, filling holes, and, in case a statistical analysis is desired, finding correspondence among each scan data. Moreover, analysis of a sagged breast is difficult to obtain, due to occlusions. In this paper, we address the problems and specific issues of using 3D scan data for the analysis and synthesis of breast models. The goal of our work is to build a breast modeler which can help both surgeons and garment designers in analyzing breast volume and surface measurements. Given enough samples of scanned breasts, our modeler can generate highly realistic breast shape, with some expected and consistent variability. Copyright © 2007 John Wiley & Sons, Ltd.

Although graphical human modeling has been a long sought subject in computer graphics, when it comes to dealing with real-time applications, it raises a number of unique requirements for both CG artists and researchers. For example, for a real-time performance it is highly desirable to lighten the geometry and optimize the runtime animation of characters wherever possible. An important need in such systems is to assist the CG artists in seamlessly transforming their modeling work from CG packages to the real-time application, without limiting their expressivity. In this paper we describe techniques by which dressed human characters are modeled, loaded into VR scenes, and simulated in real-time. In doing so, we point out challenging issues and solutions that such tasks imply at each phase of modeling and simulation. In particular, we turn our attention to what we consider as key technical components: body and clothes. Case studies show that our modelers successfully meet the need of real-time requirements over a wide variety of scenarios.

We present two complementary methodologies to automatically generate 3D surfaces of human bodies including internal bones. The first methodology is a data-driven approach and uses a database of human bodies in conjunction with a learning model synthesizer to create 3D surfaces subject to desired shape and size. In the second methodology, we use medical scanning devices to capture the internal anatomy of the individual and reconstruct 3D surfaces of bones. The final result is an animatable human body including skin external surface as well as the internal skeleton.

Summary form only given. Appropriate combinations of statistical and geometric methods permit modeling and estimation of complex objects. This paper investigates such combinations in the modeling of human body models in computer graphics. By concentrating on such a particular class of objects, we show how example based techniques together with interpolation ones can exploit existing data and knowledge about body shape and size modelling. Graphical modeling of human bodies obviously has applications in the entertainment industry for character animation and simulations involving people. Other applications range from virtual reality to ergonomic and manufacturing studies.

In this paper, we present an automatic, runtime modeler for modeling realistic, animatable human bodies. A user can generate a new model or modify an existing one simply by inputting a number of sizing parameters. We approach the problem by forming deformation functions that are devoted to the generation of appropriate shape and proportion of the body geometry by taking the parameters as input. Starting from a number of 3D scanned data of human body models as examples, we derive these functions by using radial basis interpolation. A prerequisite of such formulation is to have correspondence among example models in the database. We obtain the correspondence by fitting a template onto each scanned data. Throughout the paper, body geometry is considered to have two distinct entities, namely rigid and elastic component of the deformation. The rigid deformation is represented by the corresponding joint parameters, which will determine the linear approximation of the physique. The elastic deformation is essentially vertex displacements, which, when added to the rigid deformation, depicts the detail shape of the body. Having these interpolators formulated, the runtime modeling can be reduced to the function evaluation and application of the evaluated results to the template model. We demonstrate our method by applying different parameters to generate a wide range of different body models.

Simulating virtual environments populated with virtual but realistic crowds requires dozens of different face and body geometries. We present GENESIS (GENeration of E-population based on Statistical Information), an application that allows automatic generation of desired population models. The aim of this application is to generate any population group that is statistically calculated to satisfy given properties. We divide the population generation module into the face part and the body part. Each of them is based on a database, which is an organization of information collected and processed from the real population dataset. Upon the specification of the population parameters by the user, GENESIS transforms them onto a number Of queries to the underlying databases, and process the results from the database in a statistical manner, allowing dynamic generation of desired number of individuals while satisfying the given properties of on the overall population.

Presents the VPARK (Virtual Park) system, which includes a networked virtual environment (NVE) system called W-VLNET and an "attraction building system" that is able to create and modify the attractions used in the NVE. Both systems have been developed in the Windows NT environment. The paper outlines the techniques for communication, scene management, facial and body animation, and general user interaction modules. The use of VRML97 and MPEG-4 SHNC is overviewed for the purpose of outlining the compatability of the system with other similar virtual reality systems. The software provides realistic virtual actors as well as sets of high-level actions that are applicable to them in real-time. Related issues on obtaining actor models and animating them in real time are presented. The creation process of an attraction incorporates assembling animation units through a timeline. Using this software, the users are able to bring their own scenario-based applications into a shared virtual environment.

Presents our work on a level-of-detail (LoD) technique for human-like face models in virtual environments. Conventional LoD techniques have been adapted to allow facial animation on simplified geometric models. This includes the optimization of both geometric and animation parameters. Simplified models are generated in a region-based manner, considering the mobility of each region. The animation process is decomposed into two sub-processes, and each step is optimized. In the MPA (minimum perceptible action) level optimization, a hierarchical structure is devised for the multi-level animation model. The deformation level is simplified by reducing the number of control points. At run-time, the animation level is selected in combination with viewpoint information at the geometric level.