Kun Zhou

Hangyu Guo, Kun Zhou, Wayne Xin Zhao et al.

Although pre-trained language models~(PLMs) have shown impressive performance by text-only self-supervised training, they are found lack of visual semantics or commonsense. Existing solutions often rely on explicit images for visual knowledge augmentation (requiring time-consuming retrieval or generation), and they also conduct the augmentation for the whole input text, without considering whether it is actually needed in specific inputs or tasks. To address these issues, we propose a novel \textbf{V}isually-\textbf{A}ugmented fine-tuning approach that can be generally applied to various PLMs or NLP tasks, \textbf{W}ithout using any retrieved or generated \textbf{I}mages, namely \textbf{VAWI}. Experimental results show that our approach can consistently improve the performance of BERT, RoBERTa, BART, and T5 at different scales, and outperform several competitive baselines on ten tasks. Our codes and data are publicly available at~\url{https://github.com/RUCAIBox/VAWI}.

Kun Zhou, Wenbo Li, Yi Wang et al.

Neural radiance fields (NeRF) show great success in novel view synthesis. However, in real-world scenes, recovering high-quality details from the source images is still challenging for the existing NeRF-based approaches, due to the potential imperfect calibration information and scene representation inaccuracy. Even with high-quality training frames, the synthetic novel views produced by NeRF models still suffer from notable rendering artifacts, such as noise, blur, etc. Towards to improve the synthesis quality of NeRF-based approaches, we propose NeRFLiX, a general NeRF-agnostic restorer paradigm by learning a degradation-driven inter-viewpoint mixer. Specially, we design a NeRF-style degradation modeling approach and construct large-scale training data, enabling the possibility of effectively removing NeRF-native rendering artifacts for existing deep neural networks. Moreover, beyond the degradation removal, we propose an inter-viewpoint aggregation framework that is able to fuse highly related high-quality training images, pushing the performance of cutting-edge NeRF models to entirely new levels and producing highly photo-realistic synthetic views.

Kun Zhou, Berrak Sisman, Rajib Rana et al.

Emotional speech synthesis aims to synthesize human voices with various emotional effects. The current studies are mostly focused on imitating an averaged style belonging to a specific emotion type. In this paper, we seek to generate speech with a mixture of emotions at run-time. We propose a novel formulation that measures the relative difference between the speech samples of different emotions. We then incorporate our formulation into a sequence-to-sequence emotional text-to-speech framework. During the training, the framework does not only explicitly characterize emotion styles, but also explores the ordinal nature of emotions by quantifying the differences with other emotions. At run-time, we control the model to produce the desired emotion mixture by manually defining an emotion attribute vector. The objective and subjective evaluations have validated the effectiveness of the proposed framework. To our best knowledge, this research is the first study on modelling, synthesizing, and evaluating mixed emotions in speech.

Yunfeng Diao, He Wang, Tianjia Shao et al.

Human Activity Recognition (HAR) has been employed in a wide range of applications, e.g. self-driving cars, where safety and lives are at stake. Recently, the robustness of skeleton-based HAR methods have been questioned due to their vulnerability to adversarial attacks. However, the proposed attacks require the full-knowledge of the attacked classifier, which is overly restrictive. In this paper, we show such threats indeed exist, even when the attacker only has access to the input/output of the model. To this end, we propose the very first black-box adversarial attack approach in skeleton-based HAR called BASAR. BASAR explores the interplay between the classification boundary and the natural motion manifold. To our best knowledge, this is the first time data manifold is introduced in adversarial attacks on time series. Via BASAR, we find on-manifold adversarial samples are extremely deceitful and rather common in skeletal motions, in contrast to the common belief that adversarial samples only exist off-manifold. Through exhaustive evaluation, we show that BASAR can deliver successful attacks across classifiers, datasets, and attack modes. By attack, BASAR helps identify the potential causes of the model vulnerability and provides insights on possible improvements. Finally, to mitigate the newly identified threat, we propose a new adversarial training approach by leveraging the sophisticated distributions of on/off-manifold adversarial samples, called mixed manifold-based adversarial training (MMAT). MMAT can successfully help defend against adversarial attacks without compromising classification accuracy.

Kun Zhou, Wenbo Li, Nianjuan Jiang et al.

Neural radiance fields (NeRF) have shown great success in novel view synthesis. However, recovering high-quality details from real-world scenes is still challenging for the existing NeRF-based approaches, due to the potential imperfect calibration information and scene representation inaccuracy. Even with high-quality training frames, the synthetic novel views produced by NeRF models still suffer from notable rendering artifacts, such as noise and blur. To address this, we propose NeRFLiX, a general NeRF-agnostic restorer paradigm that learns a degradation-driven inter-viewpoint mixer. Specially, we design a NeRF-style degradation modeling approach and construct large-scale training data, enabling the possibility of effectively removing NeRF-native rendering artifacts for deep neural networks. Moreover, beyond the degradation removal, we propose an inter-viewpoint aggregation framework that fuses highly related high-quality training images, pushing the performance of cutting-edge NeRF models to entirely new levels and producing highly photo-realistic synthetic views. Based on this paradigm, we further present NeRFLiX++ with a stronger two-stage NeRF degradation simulator and a faster inter-viewpoint mixer, achieving superior performance with significantly improved computational efficiency. Notably, NeRFLiX++ is capable of restoring photo-realistic ultra-high-resolution outputs from noisy low-resolution NeRF-rendered views. Extensive experiments demonstrate the excellent restoration ability of NeRFLiX++ on various novel view synthesis benchmarks.

Beijia Chen, Yuefan Shen, Hongbo Fu et al.

In this paper, we present NeuralReshaper, a novel method for semantic reshaping of human bodies in single images using deep generative networks. To achieve globally coherent reshaping effects, our approach follows a fit-then-reshape pipeline, which first fits a parametric 3D human model to a source human image and then reshapes the fitted 3D model with respect to user-specified semantic attributes. Previous methods rely on image warping to transfer 3D reshaping effects to the entire image domain and thus often cause distortions in both foreground and background. In contrast, we resort to generative adversarial nets conditioned on the source image and a 2D warping field induced by the reshaped 3D model, to achieve more realistic reshaping results. Specifically, we separately encode the foreground and background information in the source image using a two-headed UNet-like generator, and guide the information flow from the foreground branch to the background branch via feature space warping. Furthermore, to deal with the lack-of-data problem that no paired data exist (i.e., the same human bodies in varying shapes), we introduce a novel self-supervised strategy to train our network. Unlike previous methods that often require manual efforts to correct undesirable artifacts caused by incorrect body-to-image fitting, our method is fully automatic. Extensive experiments on both indoor and outdoor datasets demonstrate the superiority of our method over previous approaches.

Jiawei Lu, Yingpeng Zhang, Zengjun Zhao et al.

Large-scale text-guided image diffusion models have shown astonishing results in text-to-image (T2I) generation. However, applying these models to synthesize textures for 3D geometries remains challenging due to the domain gap between 2D images and textures on a 3D surface. Early works that used a projecting-and-inpainting approach managed to preserve generation diversity but often resulted in noticeable artifacts and style inconsistencies. While recent methods have attempted to address these inconsistencies, they often introduce other issues, such as blurring, over-saturation, or over-smoothing. To overcome these challenges, we propose a novel text-to-texture synthesis framework that leverages pretrained diffusion models. We first introduce a local attention reweighing mechanism in the self-attention layers to guide the model in concentrating on spatial-correlated patches across different views, thereby enhancing local details while preserving cross-view consistency. Additionally, we propose a novel latent space merge pipeline, which further ensures consistency across different viewpoints without sacrificing too much diversity. Our method significantly outperforms existing state-of-the-art techniques regarding texture consistency and visual quality, while delivering results much faster than distillation-based methods. Importantly, our framework does not require additional training or fine-tuning, making it highly adaptable to a wide range of models available on public platforms.

Xiaohe Ma, Yaxin Yu, Hongzhi Wu et al.

We present a novel framework to efficiently acquire near-planar anisotropic reflectance in a pixel-independent fashion, using a deep gated mixtureof-experts. While existing work employs a unified network to handle all possible input, our network automatically learns to condition on the input for enhanced reconstruction. We train a gating module to select one out of a number of specialized decoders for reflectance reconstruction, based on photometric measurements, essentially trading generality for quality. A common, pre-trained latent transform module is also appended to each decoder, to offset the burden of the increased number of decoders. In addition, the illumination conditions during acquisition can be jointly optimized. The effectiveness of our framework is validated on a wide variety of challenging samples using a near-field lightstage. Compared with the state-of-the-art technique, our results are improved at the same input bandwidth, and our bandwidth can be reduced to about 1/3 for equal-quality results.

Youyi Zhan, Tianjia Shao, Yin Yang et al.

Many works have succeeded in reconstructing Gaussian human avatars from multi-view videos. However, they either struggle to capture pose-dependent appearance details with a single MLP, or rely on a computationally intensive neural network to reconstruct high-fidelity appearance but with rendering performance degraded to non-real-time. We propose a novel Gaussian human avatar representation that can reconstruct high-fidelity pose-dependence appearance with details and meanwhile can be rendered in real time. Our Gaussian avatar is empowered by spatially distributed MLPs which are explicitly located on different positions on human body. The parameters stored in each Gaussian are obtained by interpolating from the outputs of its nearby MLPs based on their distances. To avoid undesired smooth Gaussian property changing during interpolation, for each Gaussian we define a set of Gaussian offset basis, and a linear combination of basis represents the Gaussian property offsets relative to the neutral properties. Then we propose to let the MLPs output a set of coefficients corresponding to the basis. In this way, although Gaussian coefficients are derived from interpolation and change smoothly, the Gaussian offset basis is learned freely without constraints. The smoothly varying coefficients combined with freely learned basis can still produce distinctly different Gaussian property offsets, allowing the ability to learn high-frequency spatial signals. We further use control points to constrain the Gaussians distributed on a surface layer rather than allowing them to be irregularly distributed inside the body, to help the human avatar generalize better when animated under novel poses. Compared to the state-of-the-art method, our method achieves better appearance quality with finer details while the rendering speed is significantly faster under novel views and novel poses.

Zhexi Peng, Kun Zhou, Tianjia Shao

While recent Gaussian-based SLAM methods achieve photorealistic reconstruction from RGB-D data, their computational performance remains a critical bottleneck. State-of-the-art techniques operate at less than 20 fps, significantly lagging behind geometry-centric approaches like KinectFusion (hundreds of fps). This limitation stems from the heavy computational burden: modeling scenes requires numerous Gaussians and complex iterative optimization to fit RGB-D data, where insufficient Gaussian counts or optimization iterations cause severe quality degradation. To address this, we propose a Gaussian-SDF hybrid representation, combining a colorized Signed Distance Field (SDF) for smooth geometry and appearance with 3D Gaussians to capture underrepresented details. The SDF is efficiently constructed via RGB-D fusion (as in geometry-centric methods), while Gaussians undergo iterative optimization. Our representation enables drastic Gaussian reduction (50% fewer) by avoiding full-scene Gaussian modeling, and efficient Gaussian optimization (75% fewer iterations) through targeted appearance refinement. Building upon this representation, we develop GPS-SLAM (Gaussian-Plus-SDF SLAM), a real-time 3D reconstruction system achieving over 150 fps on real-world Azure Kinect sequences -- delivering an order-of-magnitude speedup over state-of-the-art techniques while maintaining comparable reconstruction quality. We will release the source code and data to facilitate future research.

Jilin Tang, Yi Yuan, Tianjia Shao et al.

In this paper we tackle the problem of pose guided person image generation, which aims to transfer a person image from the source pose to a novel target pose while maintaining the source appearance. Given the inefficiency of standard CNNs in handling large spatial transformation, we propose a structure-aware flow based method for high-quality person image generation. Specifically, instead of learning the complex overall pose changes of human body, we decompose the human body into different semantic parts (e.g., head, torso, and legs) and apply different networks to predict the flow fields for these parts separately. Moreover, we carefully design the network modules to effectively capture the local and global semantic correlations of features within and among the human parts respectively. Extensive experimental results show that our method can generate high-quality results under large pose discrepancy and outperforms state-of-the-art methods in both qualitative and quantitative comparisons.

Kun Zhou, Xiaoguang Han, Nianjuan Jiang et al.

Estimating 3D human pose from a single image is a challenging task. This work attempts to address the uncertainty of lifting the detected 2D joints to the 3D space by introducing an intermediate state-Part-Centric Heatmap Triplets (HEMlets), which shortens the gap between the 2D observation and the 3D interpretation. The HEMlets utilize three joint-heatmaps to represent the relative depth information of the end-joints for each skeletal body part. In our approach, a Convolutional Network (ConvNet) is first trained to predict HEMlets from the input image, followed by a volumetric joint-heatmap regression. We leverage on the integral operation to extract the joint locations from the volumetric heatmaps, guaranteeing end-to-end learning. Despite the simplicity of the network design, the quantitative comparisons show a significant performance improvement over the best-of-grade methods (e.g. $20\%$ on Human3.6M). The proposed method naturally supports training with "in-the-wild" images, where only weakly-annotated relative depth information of skeletal joints is available. This further improves the generalization ability of our model, as validated by qualitative comparisons on outdoor images. Leveraging the strength of the HEMlets pose estimation, we further design and append a shallow yet effective network module to regress the SMPL parameters of the body pose and shape. We term the entire HEMlets-based human pose and shape recovery pipeline HEMlets PoSh. Extensive quantitative and qualitative experiments on the existing human body recovery benchmarks justify the state-of-the-art results obtained with our HEMlets PoSh approach.

Kun Zhou, Xiaoguang Han, Nianjuan Jiang et al.

Estimating 3D human pose from a single image is a challenging task. This work attempts to address the uncertainty of lifting the detected 2D joints to the 3D space by introducing an intermediate state - Part-Centric Heatmap Triplets (HEMlets), which shortens the gap between the 2D observation and the 3D interpretation. The HEMlets utilize three joint-heatmaps to represent the relative depth information of the end-joints for each skeletal body part. In our approach, a Convolutional Network (ConvNet) is first trained to predict HEMlests from the input image, followed by a volumetric joint-heatmap regression. We leverage on the integral operation to extract the joint locations from the volumetric heatmaps, guaranteeing end-to-end learning. Despite the simplicity of the network design, the quantitative comparisons show a significant performance improvement over the best-of-grade method (by 20% on Human3.6M). The proposed method naturally supports training with "in-the-wild" images, where only weakly-annotated relative depth information of skeletal joints is available. This further improves the generalization ability of our model, as validated by qualitative comparisons on outdoor images.

Kun Zhou, Jinmiao Cai, Yao Li et al.

In this paper, a novel deep-learning based framework is proposed to infer 3D human poses from a single image. Specifically, a two-phase approach is developed. We firstly utilize a generator with two branches for the extraction of explicit and implicit depth information respectively. During the training process, an adversarial scheme is also employed to further improve the performance. The implicit and explicit depth information with the estimated 2D joints generated by a widely used estimator, in the second step, are together fed into a deep 3D pose regressor for the final pose generation. Our method achieves MPJPE of 58.68mm on the ECCV2018 3D Human Pose Estimation Challenge.

Yulong Shi, Xiaoguang Han, Nianjuan Jiang et al.

Although significant advances have been made in the area of human poses estimation from images using deep Convolutional Neural Network (ConvNet), it remains a big challenge to perform 3D pose inference in-the-wild. This is due to the difficulty to obtain 3D pose groundtruth for outdoor environments. In this paper, we propose a novel framework to tackle this problem by exploiting the information of each bone indicating if it is forward or backward with respect to the view of the camera(we term it Forwardor-Backward Information abbreviated as FBI). Our method firstly trains a ConvNet with two branches which maps an image of a human to both the 2D joint locations and the FBI of bones. These information is further fed into a deep regression network to predict the 3D positions of joints. To support the training, we also develop an annotation user interface and labeled such FBI for around 12K in-the-wild images which are randomly selected from MPII (a public dataset of 2D pose annotation). Our experimental results on the standard benchmarks demonstrate that our approach outperforms state-of-the-art methods both qualitatively and quantitatively.