Ziyu Wang

Zhihui Lin, Tianyu Yang, Maomao Li et al.

Matching-based methods, especially those based on space-time memory, are significantly ahead of other solutions in semi-supervised video object segmentation (VOS). However, continuously growing and redundant template features lead to an inefficient inference. To alleviate this, we propose a novel Sequential Weighted Expectation-Maximization (SWEM) network to greatly reduce the redundancy of memory features. Different from the previous methods which only detect feature redundancy between frames, SWEM merges both intra-frame and inter-frame similar features by leveraging the sequential weighted EM algorithm. Further, adaptive weights for frame features endow SWEM with the flexibility to represent hard samples, improving the discrimination of templates. Besides, the proposed method maintains a fixed number of template features in memory, which ensures the stable inference complexity of the VOS system. Extensive experiments on commonly used DAVIS and YouTube-VOS datasets verify the high efficiency (36 FPS) and high performance (84.3\% $\mathcal{J}\&\mathcal{F}$ on DAVIS 2017 validation dataset) of SWEM. Code is available at: https://github.com/lmm077/SWEM.

Ziyu Wang, Mike Zheng Shou, Mengmi Zhang

Learning object-centric representations from complex natural environments enables both humans and machines with reasoning abilities from low-level perceptual features. To capture compositional entities of the scene, we proposed cyclic walks between perceptual features extracted from vision transformers and object entities. First, a slot-attention module interfaces with these perceptual features and produces a finite set of slot representations. These slots can bind to any object entities in the scene via inter-slot competitions for attention. Next, we establish entity-feature correspondence with cyclic walks along high transition probability based on the pairwise similarity between perceptual features (aka "parts") and slot-binded object representations (aka "whole"). The whole is greater than its parts and the parts constitute the whole. The part-whole interactions form cycle consistencies, as supervisory signals, to train the slot-attention module. Our rigorous experiments on \textit{seven} image datasets in \textit{three} \textit{unsupervised} tasks demonstrate that the networks trained with our cyclic walks can disentangle foregrounds and backgrounds, discover objects, and segment semantic objects in complex scenes. In contrast to object-centric models attached with a decoder for the pixel-level or feature-level reconstructions, our cyclic walks provide strong learning signals, avoiding computation overheads and enhancing memory efficiency. Our source code and data are available at: \href{https://github.com/ZhangLab-DeepNeuroCogLab/Parts-Whole-Object-Centric-Learning/}{link}.

Liao Wang, Qiang Hu, Qihan He et al.

The success of the Neural Radiance Fields (NeRFs) for modeling and free-view rendering static objects has inspired numerous attempts on dynamic scenes. Current techniques that utilize neural rendering for facilitating free-view videos (FVVs) are restricted to either offline rendering or are capable of processing only brief sequences with minimal motion. In this paper, we present a novel technique, Residual Radiance Field or ReRF, as a highly compact neural representation to achieve real-time FVV rendering on long-duration dynamic scenes. ReRF explicitly models the residual information between adjacent timestamps in the spatial-temporal feature space, with a global coordinate-based tiny MLP as the feature decoder. Specifically, ReRF employs a compact motion grid along with a residual feature grid to exploit inter-frame feature similarities. We show such a strategy can handle large motions without sacrificing quality. We further present a sequential training scheme to maintain the smoothness and the sparsity of the motion/residual grids. Based on ReRF, we design a special FVV codec that achieves three orders of magnitudes compression rate and provides a companion ReRF player to support online streaming of long-duration FVVs of dynamic scenes. Extensive experiments demonstrate the effectiveness of ReRF for compactly representing dynamic radiance fields, enabling an unprecedented free-viewpoint viewing experience in speed and quality.

Ziyu Wang, Yu Deng, Jiaolong Yang et al.

3D-aware generative models have demonstrated their superb performance to generate 3D neural radiance fields (NeRF) from a collection of monocular 2D images even for topology-varying object categories. However, these methods still lack the capability to separately control the shape and appearance of the objects in the generated radiance fields. In this paper, we propose a generative model for synthesizing radiance fields of topology-varying objects with disentangled shape and appearance variations. Our method generates deformable radiance fields, which builds the dense correspondence between the density fields of the objects and encodes their appearances in a shared template field. Our disentanglement is achieved in an unsupervised manner without introducing extra labels to previous 3D-aware GAN training. We also develop an effective image inversion scheme for reconstructing the radiance field of an object in a real monocular image and manipulating its shape and appearance. Experiments show that our method can successfully learn the generative model from unstructured monocular images and well disentangle the shape and appearance for objects (e.g., chairs) with large topological variance. The model trained on synthetic data can faithfully reconstruct the real object in a given single image and achieve high-quality texture and shape editing results.

Ziyu Wang, Wei Yang, Junming Cao et al.

Existing neural reconstruction schemes such as Neural Radiance Field (NeRF) are largely focused on modeling opaque objects. We present a novel neural refractive field(NeReF) to recover wavefront of transparent fluids by simultaneously estimating the surface position and normal of the fluid front. Unlike prior arts that treat the reconstruction target as a single layer of the surface, NeReF is specifically formulated to recover a volumetric normal field with its corresponding density field. A query ray will be refracted by NeReF according to its accumulated refractive point and normal, and we employ the correspondences and uniqueness of refracted ray for NeReF optimization. We show NeReF, as a global optimization scheme, can more robustly tackle refraction distortions detrimental to traditional methods for correspondence matching. Furthermore, the continuous NeReF representation of wavefront enables view synthesis as well as normal integration. We validate our approach on both synthetic and real data and show it is particularly suitable for sparse multi-view acquisition. We hence build a small light field array and experiment on various surface shapes to demonstrate high fidelity NeReF reconstruction.

Yang Qu, Yutian Qin, Lecheng Chao et al.

The relationship between perceptual loudness and physical attributes of sound is an important subject in both computer music and psychoacoustics. Early studies of "equal-loudness contour" can trace back to the 1920s and the measured loudness with respect to intensity and frequency has been revised many times since then. However, most studies merely focus on synthesized sound, and the induced theories on natural tones with complex timbre have rarely been justified. To this end, we investigate both theory and applications of natural-tone loudness perception in this paper via modeling piano tone. The theory part contains: 1) an accurate measurement of piano-tone equal-loudness contour of pitches, and 2) a machine-learning model capable of inferring loudness purely based on spectral features trained on human subject measurements. As for the application, we apply our theory to piano control transfer, in which we adjust the MIDI velocities on two different player pianos (in different acoustic environments) to achieve the same perceptual effect. Experiments show that both our theoretical loudness modeling and the corresponding performance control transfer algorithm significantly outperform their baselines.

Ziyu Wang, Wenhao Jiang, Yiming Zhu et al.

Recently, MLP-like vision models have achieved promising performances on mainstream visual recognition tasks. In contrast with vision transformers and CNNs, the success of MLP-like models shows that simple information fusion operations among tokens and channels can yield a good representation power for deep recognition models. However, existing MLP-like models fuse tokens through static fusion operations, lacking adaptability to the contents of the tokens to be mixed. Thus, customary information fusion procedures are not effective enough. To this end, this paper presents an efficient MLP-like network architecture, dubbed DynaMixer, resorting to dynamic information fusion. Critically, we propose a procedure, on which the DynaMixer model relies, to dynamically generate mixing matrices by leveraging the contents of all the tokens to be mixed. To reduce the time complexity and improve the robustness, a dimensionality reduction technique and a multi-segment fusion mechanism are adopted. Our proposed DynaMixer model (97M parameters) achieves 84.3\% top-1 accuracy on the ImageNet-1K dataset without extra training data, performing favorably against the state-of-the-art vision MLP models. When the number of parameters is reduced to 26M, it still achieves 82.7\% top-1 accuracy, surpassing the existing MLP-like models with a similar capacity. The code is available at \url{https://github.com/ziyuwwang/DynaMixer}.

Ziyu Wang, Lejun Min, Gus Xia

Recent deep music generation studies have put much emphasis on long-term generation with structures. However, we are yet to see high-quality, well-structured whole-song generation. In this paper, we make the first attempt to model a full music piece under the realization of compositional hierarchy. With a focus on symbolic representations of pop songs, we define a hierarchical language, in which each level of hierarchy focuses on the semantics and context dependency at a certain music scope. The high-level languages reveal whole-song form, phrase, and cadence, whereas the low-level languages focus on notes, chords, and their local patterns. A cascaded diffusion model is trained to model the hierarchical language, where each level is conditioned on its upper levels. Experiments and analysis show that our model is capable of generating full-piece music with recognizable global verse-chorus structure and cadences, and the music quality is higher than the baselines. Additionally, we show that the proposed model is controllable in a flexible way. By sampling from the interpretable hierarchical languages or adjusting pre-trained external representations, users can control the music flow via various features such as phrase harmonic structures, rhythmic patterns, and accompaniment texture.

Ming Cheng, Ziyi Zhou, Bowen Zhang et al.

In the landscape of spatio-temporal data analytics, effective trajectory representation learning is paramount. To bridge the gap of learning accurate representations with efficient and flexible mechanisms, we introduce Efflex, a comprehensive pipeline for transformative graph modeling and representation learning of the large-volume spatio-temporal trajectories. Efflex pioneers the incorporation of a multi-scale k-nearest neighbors (KNN) algorithm with feature fusion for graph construction, marking a leap in dimensionality reduction techniques by preserving essential data features. Moreover, the groundbreaking graph construction mechanism and the high-performance lightweight GCN increase embedding extraction speed by up to 36 times faster. We further offer Efflex in two versions, Efflex-L for scenarios demanding high accuracy, and Efflex-B for environments requiring swift data processing. Comprehensive experimentation with the Porto and Geolife datasets validates our approach, positioning Efflex as the state-of-the-art in the domain. Such enhancements in speed and accuracy highlight the versatility of Efflex, underscoring its wide-ranging potential for deployment in time-sensitive and computationally constrained applications.

Hongkai Chen, Zixin Luo, Yurun Tian et al.

Identifying robust and accurate correspondences across images is a fundamental problem in computer vision that enables various downstream tasks. Recent semi-dense matching methods emphasize the effectiveness of fusing relevant cross-view information through Transformer. In this paper, we propose several improvements upon this paradigm. Firstly, we introduce affine-based local attention to model cross-view deformations. Secondly, we present selective fusion to merge local and global messages from cross attention. Apart from network structure, we also identify the importance of enforcing spatial smoothness in loss design, which has been omitted by previous works. Based on these augmentations, our network demonstrate strong matching capacity under different settings. The full version of our network achieves state-of-the-art performance among semi-dense matching methods at a similar cost to LoFTR, while the slim version reaches LoFTR baseline's performance with only 15% computation cost and 18% parameters.

Ziyu Wang, Dejing Xu, Gus Xia et al.

Could we automatically derive the score of a piano accompaniment based on the audio of a pop song? This is the audio-to-symbolic arrangement problem we tackle in this paper. A good arrangement model should not only consider the audio content but also have prior knowledge of piano composition (so that the generation "sounds like" the audio and meanwhile maintains musicality). To this end, we contribute a cross-modal representation-learning model, which 1) extracts chord and melodic information from the audio, and 2) learns texture representation from both audio and a corrupted ground truth arrangement. We further introduce a tailored training strategy that gradually shifts the source of texture information from corrupted score to audio. In the end, the score-based texture posterior is reduced to a standard normal distribution, and only audio is needed for inference. Experiments show that our model captures major audio information and outperforms baselines in generation quality.

Liao Wang, Ziyu Wang, Pei Lin et al.

Generating ``bullet-time'' effects of human free-viewpoint videos is critical for immersive visual effects and VR/AR experience. Recent neural advances still lack the controllable and interactive bullet-time design ability for human free-viewpoint rendering, especially under the real-time, dynamic and general setting for our trajectory-aware task. To fill this gap, in this paper we propose a neural interactive bullet-time generator (iButter) for photo-realistic human free-viewpoint rendering from dense RGB streams, which enables flexible and interactive design for human bullet-time visual effects. Our iButter approach consists of a real-time preview and design stage as well as a trajectory-aware refinement stage. During preview, we propose an interactive bullet-time design approach by extending the NeRF rendering to a real-time and dynamic setting and getting rid of the tedious per-scene training. To this end, our bullet-time design stage utilizes a hybrid training set, light-weight network design and an efficient silhouette-based sampling strategy. During refinement, we introduce an efficient trajectory-aware scheme within 20 minutes, which jointly encodes the spatial, temporal consistency and semantic cues along the designed trajectory, achieving photo-realistic bullet-time viewing experience of human activities. Extensive experiments demonstrate the effectiveness of our approach for convenient interactive bullet-time design and photo-realistic human free-viewpoint video generation.

Ziyu Wang, Liao Wang, Fuqiang Zhao et al.

Photo-realistic neural reconstruction and rendering of the human portrait are critical for numerous VR/AR applications. Still, existing solutions inherently rely on multi-view capture settings, and the one-shot solution to get rid of the tedious multi-view synchronization and calibration remains extremely challenging. In this paper, we propose MirrorNeRF - a one-shot neural portrait free-viewpoint rendering approach using a catadioptric imaging system with multiple sphere mirrors and a single high-resolution digital camera, which is the first to combine neural radiance field with catadioptric imaging so as to enable one-shot photo-realistic human portrait reconstruction and rendering, in a low-cost and casual capture setting. More specifically, we propose a light-weight catadioptric system design with a sphere mirror array to enable diverse ray sampling in the continuous 3D space as well as an effective online calibration for the camera and the mirror array. Our catadioptric imaging system can be easily deployed with a low budget and the casual capture ability for convenient daily usages. We introduce a novel neural warping radiance field representation to learn a continuous displacement field that implicitly compensates for the misalignment due to our flexible system setting. We further propose a density regularization scheme to leverage the inherent geometry information from the catadioptric data in a self-supervision manner, which not only improves the training efficiency but also provides more effective density supervision for higher rendering quality. Extensive experiments demonstrate the effectiveness and robustness of our scheme to achieve one-shot photo-realistic and high-quality appearance free-viewpoint rendering for human portrait scenes.

Ziyu Wang, Shuyu Cheng, Yueru Li et al.

Score matching provides an effective approach to learning flexible unnormalized models, but its scalability is limited by the need to evaluate a second-order derivative. In this paper, we present a scalable approximation to a general family of learning objectives including score matching, by observing a new connection between these objectives and Wasserstein gradient flows. We present applications with promise in learning neural density estimators on manifolds, and training implicit variational and Wasserstein auto-encoders with a manifold-valued prior.

Ruihan Yang, Dingsu Wang, Ziyu Wang et al.

Analogy-making is a key method for computer algorithms to generate both natural and creative music pieces. In general, an analogy is made by partially transferring the music abstractions, i.e., high-level representations and their relationships, from one piece to another; however, this procedure requires disentangling music representations, which usually takes little effort for musicians but is non-trivial for computers. Three sub-problems arise: extracting latent representations from the observation, disentangling the representations so that each part has a unique semantic interpretation, and mapping the latent representations back to actual music. In this paper, we contribute an explicitly-constrained variational autoencoder (EC$^2$-VAE) as a unified solution to all three sub-problems. We focus on disentangling the pitch and rhythm representations of 8-beat music clips conditioned on chords. In producing music analogies, this model helps us to realize the imaginary situation of "what if" a piece is composed using a different pitch contour, rhythm pattern, or chord progression by borrowing the representations from other pieces. Finally, we validate the proposed disentanglement method using objective measurements and evaluate the analogy examples by a subjective study.

Tom Le Paine, Sergio Gómez Colmenarejo, Ziyu Wang et al.

Humans are experts at high-fidelity imitation -- closely mimicking a demonstration, often in one attempt. Humans use this ability to quickly solve a task instance, and to bootstrap learning of new tasks. Achieving these abilities in autonomous agents is an open problem. In this paper, we introduce an off-policy RL algorithm (MetaMimic) to narrow this gap. MetaMimic can learn both (i) policies for high-fidelity one-shot imitation of diverse novel skills, and (ii) policies that enable the agent to solve tasks more efficiently than the demonstrators. MetaMimic relies on the principle of storing all experiences in a memory and replaying these to learn massive deep neural network policies by off-policy RL. This paper introduces, to the best of our knowledge, the largest existing neural networks for deep RL and shows that larger networks with normalization are needed to achieve one-shot high-fidelity imitation on a challenging manipulation task. The results also show that both types of policy can be learned from vision, in spite of the task rewards being sparse, and without access to demonstrator actions.

Zichao Yang, Marcin Moczulski, Misha Denil et al.

The fully connected layers of a deep convolutional neural network typically contain over 90% of the network parameters, and consume the majority of the memory required to store the network parameters. Reducing the number of parameters while preserving essentially the same predictive performance is critically important for operating deep neural networks in memory constrained environments such as GPUs or embedded devices. In this paper we show how kernel methods, in particular a single Fastfood layer, can be used to replace all fully connected layers in a deep convolutional neural network. This novel Fastfood layer is also end-to-end trainable in conjunction with convolutional layers, allowing us to combine them into a new architecture, named deep fried convolutional networks, which substantially reduces the memory footprint of convolutional networks trained on MNIST and ImageNet with no drop in predictive performance.

Ziyu Wang, Frank Hutter, Masrour Zoghi et al.

Bayesian optimization techniques have been successfully applied to robotics, planning, sensor placement, recommendation, advertising, intelligent user interfaces and automatic algorithm configuration. Despite these successes, the approach is restricted to problems of moderate dimension, and several workshops on Bayesian optimization have identified its scaling to high-dimensions as one of the holy grails of the field. In this paper, we introduce a novel random embedding idea to attack this problem. The resulting Random EMbedding Bayesian Optimization (REMBO) algorithm is very simple, has important invariance properties, and applies to domains with both categorical and continuous variables. We present a thorough theoretical analysis of REMBO. Empirical results confirm that REMBO can effectively solve problems with billions of dimensions, provided the intrinsic dimensionality is low. They also show that REMBO achieves state-of-the-art performance in optimizing the 47 discrete parameters of a popular mixed integer linear programming solver.