Generating conversational gestures from speech audio is challenging due to the inherent one-to-many mapping between audio and body motions. Conventional CNNs/RNNs assume one-to-one mapping, and thus tend to predict the average of all possible target motions, resulting in plain/boring motions during inference. In order to overcome this problem, we propose a novel conditional variational autoencoder (VAE) that explicitly models one-to-many audio-to-motion mapping by splitting the cross-modal latent code into shared code and motion-specific code. The shared code mainly models the strong correlation between audio and motion (such as the synchronized audio and motion beats), while the motion-specific code captures diverse motion information independent of the audio. However, splitting the latent code into two parts poses training difficulties for the VAE model. A mapping network facilitating random sampling along with other techniques including relaxed motion loss, bicycle constraint, and diversity loss are designed to better train the VAE. Experiments on both 3D and 2D motion datasets verify that our method generates more realistic and diverse motions than state-of-the-art methods, quantitatively and qualitatively. Finally, we demonstrate that our method can be readily used to generate motion sequences with user-specified motion clips on the timeline. Code and more results are at https://jingli513.github.io/audio2gestures.
Recent research has witnessed advances in facial image editing tasks including face swapping and face reenactment. However, these methods are confined to dealing with one specific task at a time. In addition, for video facial editing, previous methods either simply apply transformations frame by frame or utilize multiple frames in a concatenated or iterative fashion, which leads to noticeable visual flickers. In this paper, we propose a unified temporally consistent facial video editing framework termed UniFaceGAN. Based on a 3D reconstruction model and a simple yet efficient dynamic training sample selection mechanism, our framework is designed to handle face swapping and face reenactment simultaneously. To enforce the temporal consistency, a novel 3D temporal loss constraint is introduced based on the barycentric coordinate interpolation. Besides, we propose a region-aware conditional normalization layer to replace the traditional AdaIN or SPADE to synthesize more context-harmonious results. Compared with the state-of-the-art facial image editing methods, our framework generates video portraits that are more photo-realistic and temporally smooth.
We present animatable neural radiance fields for detailed human avatar creation from monocular videos. Our approach extends neural radiance fields (NeRF) to the dynamic scenes with human movements via introducing explicit pose-guided deformation while learning the scene representation network. In particular, we estimate the human pose for each frame and learn a constant canonical space for the detailed human template, which enables natural shape deformation from the observation space to the canonical space under the explicit control of the pose parameters. To compensate for inaccurate pose estimation, we introduce the pose refinement strategy that updates the initial pose during the learning process, which not only helps to learn more accurate human reconstruction but also accelerates the convergence. In experiments we show that the proposed approach achieves 1) implicit human geometry and appearance reconstruction with high-quality details, 2) photo-realistic rendering of the human from arbitrary views, and 3) animation of the human with arbitrary poses.
Image-to-Image (I2I) multi-domain translation models are usually evaluated also using the quality of their semantic interpolation results. However, state-of-the-art models frequently show abrupt changes in the image appearance during interpolation, and usually perform poorly in interpolations across domains. In this paper, we propose a new training protocol based on three specific losses which help a translation network to learn a smooth and disentangled latent style space in which: 1) Both intra- and inter-domain interpolations correspond to gradual changes in the generated images and 2) The content of the source image is better preserved during the translation. Moreover, we propose a novel evaluation metric to properly measure the smoothness of latent style space of I2I translation models. The proposed method can be plugged into existing translation approaches, and our extensive experiments on different datasets show that it can significantly boost the quality of the generated images and the graduality of the interpolations.
Reconstructing a 3D hand from a single-view RGB image is challenging due to various hand configurations and depth ambiguity. To reliably reconstruct a 3D hand from a monocular image, most state-of-the-art methods heavily rely on 3D annotations at the training stage, but obtaining 3D annotations is expensive. To alleviate reliance on labeled training data, we propose S2HAND, a self-supervised 3D hand reconstruction network that can jointly estimate pose, shape, texture, and the camera viewpoint. Specifically, we obtain geometric cues from the input image through easily accessible 2D detected keypoints. To learn an accurate hand reconstruction model from these noisy geometric cues, we utilize the consistency between 2D and 3D representations and propose a set of novel losses to rationalize outputs of the neural network. For the first time, we demonstrate the feasibility of training an accurate 3D hand reconstruction network without relying on manual annotations. Our experiments show that the proposed method achieves comparable performance with recent fully-supervised methods while using fewer supervision data.
We present a fully automatic system that can produce high-fidelity, photo-realistic 3D digital human characters with a consumer RGB-D selfie camera. The system only needs the user to take a short selfie RGB-D video while rotating his/her head, and can produce a high quality reconstruction in less than 30 seconds. Our main contribution is a new facial geometry modeling and reflectance synthesis procedure that significantly improves the state-of-the-art. Specifically, given the input video a two-stage frame selection algorithm is first employed to select a few high-quality frames for reconstruction. A novel, differentiable renderer based 3D Morphable Model (3DMM) fitting method is then applied to recover facial geometries from multiview RGB-D data, which takes advantages of extensive data generation and perturbation. Our 3DMM has much larger expressive capacities than conventional 3DMM, allowing us to recover more accurate facial geometry using merely linear bases. For reflectance synthesis, we present a hybrid approach that combines parametric fitting and CNNs to synthesize high-resolution albedo/normal maps with realistic hair/pore/wrinkle details. Results show that our system can produce faithful 3D characters with extremely realistic details. Code and the constructed 3DMM is publicly available.
This paper proposes a novel pretext task to address the self-supervised video representation learning problem. Specifically, given an unlabeled video clip, we compute a series of spatio-temporal statistical summaries, such as the spatial location and dominant direction of the largest motion, the spatial location and dominant color of the largest color diversity along the temporal axis, etc. Then a neural network is built and trained to yield the statistical summaries given the video frames as inputs. In order to alleviate the learning difficulty, we employ several spatial partitioning patterns to encode rough spatial locations instead of exact spatial Cartesian coordinates. Our approach is inspired by the observation that human visual system is sensitive to rapidly changing contents in the visual field, and only needs impressions about rough spatial locations to understand the visual contents. To validate the effectiveness of the proposed approach, we conduct extensive experiments with several 3D backbone networks, i.e., C3D, 3D-ResNet and R(2+1)D. The results show that our approach outperforms the existing approaches across the three backbone networks on various downstream video analytic tasks including action recognition, video retrieval, dynamic scene recognition, and action similarity labeling. The source code is made publicly available at: https://github.com/laura-wang/video_repres_sts.
Recent research has witnessed the advances in facial image editing tasks. For video editing, however, previous methods either simply apply transformations frame by frame or utilize multiple frames in a concatenated or iterative fashion, which leads to noticeable visual flickers. In addition, these methods are confined to dealing with one specific task at a time without any extensibility. In this paper, we propose a task-agnostic temporally consistent facial video editing framework. Based on a 3D reconstruction model, our framework is designed to handle several editing tasks in a more unified and disentangled manner. The core design includes a dynamic training sample selection mechanism and a novel 3D temporal loss constraint that fully exploits both image and video datasets and enforces temporal consistency. Compared with the state-of-the-art facial image editing methods, our framework generates video portraits that are more photo-realistic and temporally smooth.
Accurate 3D reconstruction of the hand and object shape from a hand-object image is important for understanding human-object interaction as well as human daily activities. Different from bare hand pose estimation, hand-object interaction poses a strong constraint on both the hand and its manipulated object, which suggests that hand configuration may be crucial contextual information for the object, and vice versa. However, current approaches address this task by training a two-branch network to reconstruct the hand and object separately with little communication between the two branches. In this work, we propose to consider hand and object jointly in feature space and explore the reciprocity of the two branches. We extensively investigate cross-branch feature fusion architectures with MLP or LSTM units. Among the investigated architectures, a variant with LSTM units that enhances object feature with hand feature shows the best performance gain. Moreover, we employ an auxiliary depth estimation module to augment the input RGB image with the estimated depth map, which further improves the reconstruction accuracy. Experiments conducted on public datasets demonstrate that our approach significantly outperforms existing approaches in terms of the reconstruction accuracy of objects.
While deep neural networks have been shown to perform remarkably well in many machine learning tasks, labeling a large amount of ground truth data for supervised training is usually very costly to scale. Therefore, learning robust representations with unlabeled data is critical in relieving human effort and vital for many downstream tasks. Recent advances in unsupervised and self-supervised learning approaches for visual data have benefited greatly from domain knowledge. Here we are interested in a more generic unsupervised learning framework that can be easily generalized to other domains. In this paper, we propose to learn data representations with a novel type of denoising autoencoder, where the noisy input data is generated by corrupting latent clean data in the gradient domain. This can be naturally generalized to span multiple scales with a Laplacian pyramid representation of the input data. In this way, the agent learns more robust representations that exploit the underlying data structures across multiple scales. Experiments on several visual benchmarks demonstrate that better representations can be learned with the proposed approach, compared to its counterpart with single-scale corruption and other approaches. Furthermore, we also demonstrate that the learned representations perform well when transferring to other downstream vision tasks.