While time series classification and forecasting problems have been extensively studied, the cases of noisy time series data with arbitrary time sequence lengths have remained challenging. Each time series instance can be thought of as a sample realization of a noisy dynamical model, which is characterized by a continuous stochastic process. For many applications, the data are mixed and consist of several types of noisy time series sequences modeled by multiple stochastic processes, making the forecasting and classification tasks even more challenging. Instead of regressing data naively and individually to each time series type, we take a latent variable model approach using a mixtured Gaussian processes with learned spectral kernels. More specifically, we auto-assign each type of noisy time series data a signature vector called its motion code. Then, conditioned on each assigned motion code, we infer a sparse approximation of the corresponding time series using the concept of the most informative timestamps. Our unmixing classification approach involves maximizing the likelihood across all the mixed noisy time series sequences of varying lengths. This stochastic approach allows us to learn not only within a single type of noisy time series data but also across many underlying stochastic processes, giving us a way to learn multiple dynamical models in an integrated and robust manner. The different learned latent stochastic models allow us to generate specific sub-type forecasting. We provide several quantitative comparisons demonstrating the performance of our approach.

The present paper proposes an encoder-decoder model for extracting the structures of human motions represented by frame-wise discrete features in a self-supervised manner. In the proposed method, features are extracted as codes in a motion codebook without the use of human knowledge, and the relationship between these codes can be visualized on a graph. Since the codes are expected to be temporally sparse compared to the captured frame rate and can be shared by multiple sequences, the proposed network model also addresses the need for training constraints. Specifically, the model consists of self-attention layers and a vector clustering block. The attention layers contribute to finding sparse keyframes and discrete features as motion codes, which are then extracted by vector clustering. The constraints are realized as training losses so that the same motion codes can be as contiguous as possible and can be shared by multiple sequences. In addition, we propose the use of causal self-attention as a method by which to calculate attention for long sequences consisting of numerous frames. In our experiments, the sparse structures of motion codes were used to compile a graph that facilitates visualization of the relationship between the codes and the differences between sequences. We then evaluated the effectiveness of the extracted motion codes by applying them to multiple recognition tasks and found that performance levels comparable to task-optimized methods could be achieved by linear probing.

Controllability, generalizability and efficiency are the major objectives of constructing face avatars represented by neural implicit field. However, existing methods have not managed to accommodate the three requirements simultaneously. They either focus on static portraits, restricting the representation ability to a specific subject, or suffer from substantial computational cost, limiting their flexibility. In this paper, we propose One-shot Talking face Avatar (OTAvatar), which constructs face avatars by a generalized controllable tri-plane rendering solution so that each personalized avatar can be constructed from only one portrait as the reference. Specifically, OTAvatar first inverts a portrait image to a motion-free identity code. Second, the identity code and a motion code are utilized to modulate an efficient CNN to generate a tri-plane formulated volume, which encodes the subject in the desired motion. Finally, volume rendering is employed to generate an image in any view. The core of our solution is a novel decoupling-by-inverting strategy that disentangles identity and motion in the latent code via optimization-based inversion. Benefiting from the efficient tri-plane representation, we achieve controllable rendering of generalized face avatar at $35$ FPS on A100. Experiments show promising performance of cross-identity reenactment on subjects out of the training set and better 3D consistency.

Videos show continuous events, yet most - if not all - video synthesis frameworks treat them discretely in time. In this work, we think of videos of what they should be - time-continuous signals, and extend the paradigm of neural representations to build a continuous-time video generator. For this, we first design continuous motion representations through the lens of positional embeddings. Then, we explore the question of training on very sparse videos and demonstrate that a good generator can be learned by using as few as 2 frames per clip. After that, we rethink the traditional image and video discriminators pair and propose to use a single hypernetwork-based one. This decreases the training cost and provides richer learning signal to the generator, making it possible to train directly on 1024$^2$ videos for the first time. We build our model on top of StyleGAN2 and it is just 5% more expensive to train at the same resolution while achieving almost the same image quality. Moreover, our latent space features similar properties, enabling spatial manipulations that our method can propagate in time. We can generate arbitrarily long videos at arbitrary high frame rate, while prior work struggles to generate even 64 frames at a fixed rate. Our model achieves state-of-the-art results on four modern 256$^2$ video synthesis benchmarks and one 1024$^2$ resolution one. Videos and the source code are available at the project website: https://universome.github.io/stylegan-v.

In this work, we propose a motion embedding strategy known as motion codes, which is a vectorized representation of motions based on a manipulation's salient mechanical attributes. These motion codes provide a robust motion representation, and they are obtained using a hierarchy of features called the motion taxonomy. We developed and trained a deep neural network model that combines visual and semantic features to identify the features found in our motion taxonomy to embed or annotate videos with motion codes. To demonstrate the potential of motion codes as features for machine learning tasks, we integrated the extracted features from the motion embedding model into the current state-of-the-art action recognition model. The obtained model achieved higher accuracy than the baseline model for the verb classification task on egocentric videos from the EPIC-KITCHENS dataset.