Search

This paper introduces ELMO, a real-time upsampling motion capture framework designed for a single LiDAR sensor. Modeled as a conditional autoregressive transformer-based upsampling motion generator, ELMO achieves 60 fps motion capture from a 20 fps LiDAR point cloud sequence. The key feature of ELMO is the coupling of the self-attention mechanism with thoughtfully designed embedding modules for motion and point clouds, significantly elevating the motion quality. To facilitate accurate motion capture, we develop a one-time skeleton calibration model capable of predicting user skeleton offsets from a single-frame point cloud. Additionally, we introduce a novel data augmentation technique utilizing a LiDAR simulator, which enhances global root tracking to improve environmental understanding. To demonstrate the effectiveness of our method, we compare ELMO with state-of-the-art methods in both image-based and point cloud-based motion capture. We further conduct an ablation study to validate our design principles. ELMO's fast inference time makes it well-suited for real-time applications, exemplified in our demo video featuring live streaming and interactive gaming scenarios. Furthermore, we contribute a high-quality LiDAR-mocap synchronized dataset comprising 20 different subjects performing a range of motions, which can serve as a valuable resource for future research. The dataset and evaluation code are available at {\blue \url{https://movin3d.github.io/ELMO_SIGASIA2024/}}, Comment: published at ACM Transactions on Graphics (Proc. SIGGRAPH ASIA), 2024

Full-body avatar presence is crucial for immersive social and environmental interactions in digital reality. However, current devices only provide three six degrees of freedom (DOF) poses from the headset and two controllers (i.e. three-point trackers). Because it is a highly under-constrained problem, inferring full-body pose from these inputs is challenging, especially when supporting the full range of body proportions and use cases represented by the general population. In this paper, we propose a deep learning framework, DivaTrack, which outperforms existing methods when applied to diverse body sizes and activities. We augment the sparse three-point inputs with linear accelerations from Inertial Measurement Units (IMU) to improve foot contact prediction. We then condition the otherwise ambiguous lower-body pose with the predictions of foot contact and upper-body pose in a two-stage model. We further stabilize the inferred full-body pose in a wide range of configurations by learning to blend predictions that are computed in two reference frames, each of which is designed for different types of motions. We demonstrate the effectiveness of our design on a large dataset that captures 22 subjects performing challenging locomotion for three-point tracking, including lunges, hula-hooping, and sitting. As shown in a live demo using the Meta VR headset and Xsens IMUs, our method runs in real-time while accurately tracking a user's motion when they perform a diverse set of movements., Comment: accepted to Eurographics 2024

This paper presents a novel method for reconstructing 3D garment models from a single image of a posed user. Previous studies that have primarily focused on accurately reconstructing garment geometries to match the input garment image may often result in unnatural-looking garments when deformed for new poses. To overcome this limitation, our approach takes a different approach by inferring the fundamental shape of the garment through sewing patterns from a single image, rather than directly reconstructing 3D garments. Our method consists of two stages. Firstly, given a single image of a posed user, it predicts the garment image worn on a T-pose, representing the baseline form of the garment. Then, it estimates the sewing pattern parameters based on the T-pose garment image. By simulating the stitching and draping of the sewing pattern using physics simulation, we can generate 3D garments that can adaptively deform to arbitrary poses. The effectiveness of our method is validated through ablation studies on the major components and a comparison with other approaches.

Realistic reconstruction of 3D clothing from an image has wide applications, such as avatar creation and virtual try-on. This paper presents a novel framework that reconstructs the texture map for 3D garments from a single image with pose. Assuming that 3D garments are modeled by stitching 2D garment sewing patterns, our specific goal is to generate a texture image for the sewing patterns. A key component of our framework, the Texture Unwarper, infers the original texture image from the input clothing image, which exhibits warping and occlusion of texture due to the user's body shape and pose. The Texture Unwarper effectively transforms between the input and output images by mapping the latent spaces of the two images. By inferring the unwarped original texture of the input garment, our method helps reconstruct 3D garment models that can show high-quality texture images realistically deformed for new poses. We validate the effectiveness of our approach through a comparison with other methods and ablation studies.

Transforming neutral, characterless input motions to embody the distinct style of a notable character in real time is highly compelling for character animation. This paper introduces MOCHA, a novel online motion characterization framework that transfers both motion styles and body proportions from a target character to an input source motion. MOCHA begins by encoding the input motion into a motion feature that structures the body part topology and captures motion dependencies for effective characterization. Central to our framework is the Neural Context Matcher, which generates a motion feature for the target character with the most similar context to the input motion feature. The conditioned autoregressive model of the Neural Context Matcher can produce temporally coherent character features in each time frame. To generate the final characterized pose, our Characterizer network incorporates the characteristic aspects of the target motion feature into the input motion feature while preserving its context. This is achieved through a transformer model that introduces the adaptive instance normalization and context mapping-based cross-attention, effectively injecting the character feature into the source feature. We validate the performance of our framework through comparisons with prior work and an ablation study. Our framework can easily accommodate various applications, including characterization with only sparse input and real-time characterization. Additionally, we contribute a high-quality motion dataset comprising six different characters performing a range of motions, which can serve as a valuable resource for future research., Comment: presented at Siggraph Asia 2023

Recent advancements in technology have brought forth new forms of interactive applications, such as the social metaverse, where end users interact with each other through their virtual avatars. In such applications, precise full-body tracking is essential for an immersive experience and a sense of embodiment with the virtual avatar. However, current motion capture systems are not easily accessible to end users due to their high cost, the requirement for special skills to operate them, or the discomfort associated with wearable devices. In this paper, we present MOVIN, the data-driven generative method for real-time motion capture with global tracking, using a single LiDAR sensor. Our autoregressive conditional variational autoencoder (CVAE) model learns the distribution of pose variations conditioned on the given 3D point cloud from LiDAR.As a central factor for high-accuracy motion capture, we propose a novel feature encoder to learn the correlation between the historical 3D point cloud data and global, local pose features, resulting in effective learning of the pose prior. Global pose features include root translation, rotation, and foot contacts, while local features comprise joint positions and rotations. Subsequently, a pose generator takes into account the sampled latent variable along with the features from the previous frame to generate a plausible current pose. Our framework accurately predicts the performer's 3D global information and local joint details while effectively considering temporally coherent movements across frames. We demonstrate the effectiveness of our architecture through quantitative and qualitative evaluations, comparing it against state-of-the-art methods. Additionally, we implement a real-time application to showcase our method in real-world scenarios. MOVIN dataset is available at \url{https://movin3d.github.io/movin_pg2023/}.

Rapid advances in technology gradually realize immersive mixed-reality (MR) telepresence between distant spaces. This paper presents a novel visual guidance system for avatar-mediated telepresence, directing users to optimal placements that facilitate the clear transfer of gaze and pointing contexts through remote avatars in dissimilar spaces, where the spatial relationship between the remote avatar and the interaction targets may differ from that of the local user. Representing the spatial relationship between the user/avatar and interaction targets with angle-based interaction features, we assign recommendation scores of sampled local placements as their maximum feature similarity with remote placements. These scores are visualized as color-coded 2D sectors to inform the users of better placements for interaction with selected targets. In addition, virtual objects of the remote space are overlapped with the local space for the user to better understand the recommendations. We examine whether the proposed score measure agrees with the actual user perception of the partner's interaction context and find a score threshold for recommendation through user experiments in virtual reality (VR). A subsequent user study in VR investigates the effectiveness and perceptual overload of different combinations of visualizations. Finally, we conduct a user study in an MR telepresence scenario to evaluate the effectiveness of our method in real-world applications.

This paper presents Motion Puzzle, a novel motion style transfer network that advances the state-of-the-art in several important respects. The Motion Puzzle is the first that can control the motion style of individual body parts, allowing for local style editing and significantly increasing the range of stylized motions. Designed to keep the human's kinematic structure, our framework extracts style features from multiple style motions for different body parts and transfers them locally to the target body parts. Another major advantage is that it can transfer both global and local traits of motion style by integrating the adaptive instance normalization and attention modules while keeping the skeleton topology. Thus, it can capture styles exhibited by dynamic movements, such as flapping and staggering, significantly better than previous work. In addition, our framework allows for arbitrary motion style transfer without datasets with style labeling or motion pairing, making many publicly available motion datasets available for training. Our framework can be easily integrated with motion generation frameworks to create many applications, such as real-time motion transfer. We demonstrate the advantages of our framework with a number of examples and comparisons with previous work., Comment: 16 pages

The fields of SocialVR, performance capture, and virtual try-on are often faced with a need to faithfully reproduce real garments in the virtual world. One critical task is the disentanglement of the intrinsic garment shape from deformations due to fabric properties, physical forces, and contact with the body. We propose to use a garment sewing pattern, a realistic and compact garment descriptor, to facilitate the intrinsic garment shape estimation. Another major challenge is a high diversity of shapes and designs in the domain. The most common approach for Deep Learning on 3D garments is to build specialized models for individual garments or garment types. We argue that building a unified model for various garment designs has the benefit of generalization to novel garment types, hence covering a larger design domain than individual models would. We introduce NeuralTailor, a novel architecture based on point-level attention for set regression with variable cardinality, and apply it to the task of reconstructing 2D garment sewing patterns from the 3D point could garment models. Our experiments show that NeuralTailor successfully reconstructs sewing patterns and generalizes to garment types with pattern topologies unseen during training., Comment: Updated to the version accepted to SIGGRAPH 2022 (Journal Track)

Garments are ubiquitous in both real and many of the virtual worlds. They are highly deformable objects, exhibit an immense variety of designs and shapes, and yet, most garments are created from a set of regularly shaped flat pieces. Exploration of garment structure presents a peculiar case for an object structure estimation task and might prove useful for downstream tasks of neural 3D garment modeling and reconstruction by providing strong prior on garment shapes. To facilitate research in these directions, we propose a method for generating large synthetic datasets of 3D garment designs and their sewing patterns. Our method consists of a flexible description structure for specifying parametric sewing pattern templates and the automatic generation pipeline to produce garment 3D models with little-to-none manual intervention. To add realism, the pipeline additionally creates corrupted versions of the final meshes that imitate artifacts of 3D scanning. With this pipeline, we created the first large-scale synthetic dataset of 3D garment models with their sewing patterns. The dataset contains more than 20000 garment design variations produced from 19 different base types. Seven of these garment types are specifically designed to target evaluation of the generalization across garment sewing pattern topologies., Comment: To appear in NeurIPS 2021 Datasets and Benchmarks Track

Legged robots need to be capable of walking on diverse terrain conditions. In this paper, we present a novel reinforcement learning framework for learning locomotion on non-rigid dynamic terrains. Specifically, our framework can generate quadruped locomotion on flat elastic terrain that consists of a matrix of tiles moving up and down passively when pushed by the robot's feet. A trained robot with 55cm base length can walk on terrain that can sink up to 5cm. We propose a set of observation and reward terms that enable this locomotion; in which we found that it is crucial to include the end-effector history and end-effector velocity terms into observation. We show the effectiveness of our method by training the robot with various terrain conditions.

We present a novel mixed reality (MR) telepresence system enabling a local user to interact with a remote user through full-body avatars in their own rooms. If the remote rooms have different sizes and furniture arrangements, directly applying a user's motion to an avatar leads to a mismatch of placement and deictic gesture. To overcome this problem, we retarget the placement, arm gesture, and head movement of a local user to an avatar in a remote room to preserve a local user's environment and interaction context. This allows avatars to utilize real furniture and interact with a local user and shared objects as if they were in the same room. This paper describes our system's design and implementation in detail and a set of example scenarios in the living room and office room. A qualitative user study delves into a user experience, challenges, and possible extensions of the proposed system.

With the popularization of game and VR/AR devices, there is a growing need for capturing human motion with a sparse set of tracking data. In this paper, we introduce a deep neural-network (DNN) based method for real-time prediction of the lower-body pose only from the tracking signals of the upper-body joints. Specifically, our Gated Recurrent Unit (GRU)-based recurrent architecture predicts the lower-body pose and feet contact probability from past sequence of tracking signals of the head, hands and pelvis. A major feature of our method is that the input signal is represented with the velocity of tracking signals. We show that the velocity representation better models the correlation between the upper-body and lower-body motions and increase the robustness against the diverse scales and proportions of the user body than position-orientation representations. In addition, to remove foot-skating and floating artifacts, our network predicts feet contact state, which is used to post-process the lower-body pose with inverse kinematics to preserve the contact. Our network is lightweight so as to run in real-time applications. We show the effectiveness of our method through several quantitative evaluations against other architectures and input representations, with respect to wild tracking data obtained from commercial VR devices.

Rapidly developing technologies are realizing a 3D telepresence, in which geographically separated users can interact with each other through their virtual avatars. In this paper, we present novel methods to determine the avatar's position in an indoor space to preserve the semantics of the user's position in a dissimilar indoor space with different space configurations and furniture layouts. To this end, we first perform a user survey on the preferred avatar placements for various indoor configurations and user placements, and identify a set of related attributes, including interpersonal relation, visual attention, pose, and spatial characteristics, and quantify these attributes with a set of features. By using the obtained dataset and identified features, we train a neural network that predicts the similarity between two placements. Next, we develop an avatar placement method that preserves the semantics of the placement of the remote user in a different space as much as possible. We show the effectiveness of our methods by implementing a prototype AR-based telepresence system and user evaluations., Comment: IEEE Transactions on Visualization and Computer Graphics (Early Access)

This paper presents a novel recurrent neural network-based method to construct a latent motion manifold that can represent a wide range of human motions in a long sequence. We introduce several new components to increase the spatial and temporal coverage in motion space while retaining the details of motion capture data. These include new regularization terms for the motion manifold, combination of two complementary decoders for predicting joint rotations and joint velocities, and the addition of the forward kinematics layer to consider both joint rotation and position errors. In addition, we propose a set of loss terms that improve the overall quality of the motion manifold from various aspects, such as the capability of reconstructing not only the motion but also the latent manifold vector, and the naturalness of the motion through adversarial loss. These components contribute to creating compact and versatile motion manifold that allows for creating new motions by performing random sampling and algebraic operations, such as interpolation and analogy, in the latent motion manifold., Comment: 11 pages, It will be published at Computer Graphics Forum