Changxi Zheng

Yufan Zhang, Yu Ji, Ayo Ajiboye et al.

We present a diffusion-based method for relighting dynamic portrait videos with photorealism and temporal consistency. Our method is fueled by a hybrid training dataset that consists of real-captured and rendered dynamic portrait videos with diverse subject appearances, facial motions, head poses, and known lighting conditions. Specifically, we construct an LED-based lighting system for realistic lighting emulation and high-speed video relighting data acquisition. By leveraging the image priors embedded in pre-trained video diffusion models, and using per-frame high dynamic range (HDR) environment map as lighting control, we train a high-performance generative model for realistic and identity-preserving dynamic portrait video relighting. In addition to the environment map control, our model uses a synthesized background image to enable control on the camera's exposure level and color tone. Our model can produce temporally consistent relit portrait video that looks realistic and harmonious under a provided new environment and faithfully preserve the subject's expression and fine facial features, including skin tone, wrinkles, and facial hair. Our model generalizes well to unseen data, in terms of the subject appearance, motion, and lighting condition. We perform extensive experiments on relighting in-the-wild videos with various environment maps and demonstrate practical applications on portrait photography. Results show that our method achieves state-of-the-art performance in photorealism, lighting harmony, and temporal consistency.

Honglin Chen, Rundi Wu, Eitan Grinspun et al.

Implicit Neural Spatial Representation (INSR) has emerged as an effective representation of spatially-dependent vector fields. This work explores solving time-dependent PDEs with INSR. Classical PDE solvers introduce both temporal and spatial discretizations. Common spatial discretizations include meshes and meshless point clouds, where each degree-of-freedom corresponds to a location in space. While these explicit spatial correspondences are intuitive to model and understand, these representations are not necessarily optimal for accuracy, memory usage, or adaptivity. Keeping the classical temporal discretization unchanged (e.g., explicit/implicit Euler), we explore INSR as an alternative spatial discretization, where spatial information is implicitly stored in the neural network weights. The network weights then evolve over time via time integration. Our approach does not require any training data generated by existing solvers because our approach is the solver itself. We validate our approach on various PDEs with examples involving large elastic deformations, turbulent fluids, and multi-scale phenomena. While slower to compute than traditional representations, our approach exhibits higher accuracy and lower memory consumption. Whereas classical solvers can dynamically adapt their spatial representation only by resorting to complex remeshing algorithms, our INSR approach is intrinsically adaptive. By tapping into the rich literature of classic time integrators, e.g., operator-splitting schemes, our method enables challenging simulations in contact mechanics and turbulent flows where previous neural-physics approaches struggle. Videos and codes are available on the project page: http://www.cs.columbia.edu/cg/INSR-PDE/

Rundi Wu, Changxi Zheng

Existing generative models for 3D shapes are typically trained on a large 3D dataset, often of a specific object category. In this paper, we investigate the deep generative model that learns from only a single reference 3D shape. Specifically, we present a multi-scale GAN-based model designed to capture the input shape's geometric features across a range of spatial scales. To avoid large memory and computational cost induced by operating on the 3D volume, we build our generator atop the tri-plane hybrid representation, which requires only 2D convolutions. We train our generative model on a voxel pyramid of the reference shape, without the need of any external supervision or manual annotation. Once trained, our model can generate diverse and high-quality 3D shapes possibly of different sizes and aspect ratios. The resulting shapes present variations across different scales, and at the same time retain the global structure of the reference shape. Through extensive evaluation, both qualitative and quantitative, we demonstrate that our model can generate 3D shapes of various types.

Honglin Chen, Karran Pandey, Rundi Wu et al.

Kinematic rigs provide a structured interface for articulating 3D meshes, but they lack an inherent representation of the plausible manifold of joint configurations for a given asset. Without such a pose space, stochastic sampling or manual manipulation of raw rig parameters often leads to semantic or geometric violations, such as anatomical hyperextension and non-physical self-intersections. We propose Video-informed Pose Spaces (ViPS), a feed-forward framework that discovers the latent distribution of valid articulations for auto-rigged meshes by distilling motion priors from a pretrained video diffusion model. Unlike existing methods that rely on scarce artist-authored 4D datasets, ViPS transfers generative video priors into a universal distribution over a given rig parameterization. Differentiable geometric validators applied to the skinned mesh enforce asset-specific validity without requiring manual regularizers. Our model learns a smooth, compact, and controllable pose space that supports diverse sampling, manifold projection for inverse kinematics, and temporally coherent trajectories for keyframing. Furthermore, the distilled 3D pose samples serve as precise semantic proxies for guiding video diffusion, effectively closing the loop between generative 2D priors and structured 3D kinematic control. Our evaluations show that ViPS, trained solely on video priors, matches the performance of state-of-the-art methods trained on synthetic artist-created 4D data in both plausibility and diversity. Most importantly, as a universal model, ViPS demonstrates robust zero-shot generalization to out-of-distribution species and unseen skeletal topologies.

Ruiyang Liu, Jinxu Xiang, Bowen Zhao et al.

Neural Radiance Fields (NeRF) have significantly advanced the generation of highly realistic and expressive 3D scenes. However, the task of editing NeRF, particularly in terms of geometry modification, poses a significant challenge. This issue has obstructed NeRF's wider adoption across various applications. To tackle the problem of efficiently editing neural implicit fields, we introduce Neural Impostor, a hybrid representation incorporating an explicit tetrahedral mesh alongside a multigrid implicit field designated for each tetrahedron within the explicit mesh. Our framework bridges the explicit shape manipulation and the geometric editing of implicit fields by utilizing multigrid barycentric coordinate encoding, thus offering a pragmatic solution to deform, composite, and generate neural implicit fields while maintaining a complex volumetric appearance. Furthermore, we propose a comprehensive pipeline for editing neural implicit fields based on a set of explicit geometric editing operations. We show the robustness and adaptability of our system through diverse examples and experiments, including the editing of both synthetic objects and real captured data. Finally, we demonstrate the authoring process of a hybrid synthetic-captured object utilizing a variety of editing operations, underlining the transformative potential of Neural Impostor in the field of 3D content creation and manipulation.

Xinyuan Li, Yu Ji, Yanchen Liu et al.

Reconstructing and tracking deformable surface with little or no texture has posed long-standing challenges. Fundamentally, the challenges stem from textureless surfaces lacking features for establishing cross-image correspondences. In this work, we present a novel type of markers to proactively enrich the object's surface features, and thereby ease the 3D surface reconstruction and correspondence tracking. Our markers are made of fluorescent dyes, visible only under the ultraviolet (UV) light and invisible under regular lighting condition. Leveraging the markers, we design a multi-camera system that captures surface deformation under the UV light and the visible light in a time multiplexing fashion. Under the UV light, markers on the object emerge to enrich its surface texture, allowing high-quality 3D shape reconstruction and tracking. Under the visible light, markers become invisible, allowing us to capture the object's original untouched appearance. We perform experiments on various challenging scenes, including hand gestures, facial expressions, waving cloth, and hand-object interaction. In all these cases, we demonstrate that our system is able to produce robust, high-quality 3D reconstruction and tracking.

Beichen Li, Rundi Wu, Armando Solar-Lezama et al.

Procedural materials, represented as functional node graphs, are ubiquitous in computer graphics for photorealistic material appearance design. They allow users to perform intuitive and precise editing to achieve desired visual appearances. However, creating a procedural material given an input image requires professional knowledge and significant effort. In this work, we leverage the ability to convert procedural materials into standard Python programs and fine-tune a large pre-trained vision-language model (VLM) to generate such programs from input images. To enable effective fine-tuning, we also contribute an open-source procedural material dataset and propose to perform program-level augmentation by prompting another pre-trained large language model (LLM). Through extensive evaluation, we show that our method outperforms previous methods on both synthetic and real-world examples.

Tianyuan Zhang, Hong-Xing Yu, Rundi Wu et al. · deepmind, stanford

Realistic object interactions are crucial for creating immersive virtual experiences, yet synthesizing realistic 3D object dynamics in response to novel interactions remains a significant challenge. Unlike unconditional or text-conditioned dynamics generation, action-conditioned dynamics requires perceiving the physical material properties of objects and grounding the 3D motion prediction on these properties, such as object stiffness. However, estimating physical material properties is an open problem due to the lack of material ground-truth data, as measuring these properties for real objects is highly difficult. We present PhysDreamer, a physics-based approach that endows static 3D objects with interactive dynamics by leveraging the object dynamics priors learned by video generation models. By distilling these priors, PhysDreamer enables the synthesis of realistic object responses to novel interactions, such as external forces or agent manipulations. We demonstrate our approach on diverse examples of elastic objects and evaluate the realism of the synthesized interactions through a user study. PhysDreamer takes a step towards more engaging and realistic virtual experiences by enabling static 3D objects to dynamically respond to interactive stimuli in a physically plausible manner. See our project page at https://physdreamer.github.io/.

Kaifeng Zhang, Shuo Sha, Hanxiao Jiang et al.

Robotic manipulation policies are advancing rapidly, but their direct evaluation in the real world remains costly, time-consuming, and difficult to reproduce, particularly for tasks involving deformable objects. Simulation provides a scalable and systematic alternative, yet existing simulators often fail to capture the coupled visual and physical complexity of soft-body interactions. We present a real-to-sim policy evaluation framework that constructs soft-body digital twins from real-world videos and renders robots, objects, and environments with photorealistic fidelity using 3D Gaussian Splatting. We validate our approach on representative deformable manipulation tasks, including plush toy packing, rope routing, and T-block pushing, demonstrating that simulated rollouts correlate strongly with real-world execution performance and reveal key behavioral patterns of learned policies. Our results suggest that combining physics-informed reconstruction with high-quality rendering enables reproducible, scalable, and accurate evaluation of robotic manipulation policies. Website: https://real2sim-eval.github.io/

Rundi Wu, Ruiqi Gao, Ben Poole et al.

We present CAT4D, a method for creating 4D (dynamic 3D) scenes from monocular video. CAT4D leverages a multi-view video diffusion model trained on a diverse combination of datasets to enable novel view synthesis at any specified camera poses and timestamps. Combined with a novel sampling approach, this model can transform a single monocular video into a multi-view video, enabling robust 4D reconstruction via optimization of a deformable 3D Gaussian representation. We demonstrate competitive performance on novel view synthesis and dynamic scene reconstruction benchmarks, and highlight the creative capabilities for 4D scene generation from real or generated videos. See our project page for results and interactive demos: https://cat-4d.github.io/.

Basile Van Hoorick, Rundi Wu, Ege Ozguroglu et al.

Accurate reconstruction of complex dynamic scenes from just a single viewpoint continues to be a challenging task in computer vision. Current dynamic novel view synthesis methods typically require videos from many different camera viewpoints, necessitating careful recording setups, and significantly restricting their utility in the wild as well as in terms of embodied AI applications. In this paper, we propose $\textbf{GCD}$, a controllable monocular dynamic view synthesis pipeline that leverages large-scale diffusion priors to, given a video of any scene, generate a synchronous video from any other chosen perspective, conditioned on a set of relative camera pose parameters. Our model does not require depth as input, and does not explicitly model 3D scene geometry, instead performing end-to-end video-to-video translation in order to achieve its goal efficiently. Despite being trained on synthetic multi-view video data only, zero-shot real-world generalization experiments show promising results in multiple domains, including robotics, object permanence, and driving environments. We believe our framework can potentially unlock powerful applications in rich dynamic scene understanding, perception for robotics, and interactive 3D video viewing experiences for virtual reality.

Mutian Tong, Rundi Wu, Changxi Zheng

Indoor lighting estimation from a single image or video remains a challenge due to its highly ill-posed nature, especially when the lighting condition of the scene varies spatially and temporally. We propose a method that estimates from an input video a continuous light field describing the spatiotemporally varying lighting of the scene. We leverage 2D diffusion priors for optimizing such light field represented as a MLP. To enable zero-shot generalization to in-the-wild scenes, we fine-tune a pre-trained image diffusion model to predict lighting at multiple locations by jointly inpainting multiple chrome balls as light probes. We evaluate our method on indoor lighting estimation from a single image or video and show superior performance over compared baselines. Most importantly, we highlight results on spatiotemporally consistent lighting estimation from in-the-wild videos, which is rarely demonstrated in previous works.

Rundi Wu, Ruoshi Liu, Carl Vondrick et al.

Synthesizing novel 3D models that resemble the input example has long been pursued by graphics artists and machine learning researchers. In this paper, we present Sin3DM, a diffusion model that learns the internal patch distribution from a single 3D textured shape and generates high-quality variations with fine geometry and texture details. Training a diffusion model directly in 3D would induce large memory and computational cost. Therefore, we first compress the input into a lower-dimensional latent space and then train a diffusion model on it. Specifically, we encode the input 3D textured shape into triplane feature maps that represent the signed distance and texture fields of the input. The denoising network of our diffusion model has a limited receptive field to avoid overfitting, and uses triplane-aware 2D convolution blocks to improve the result quality. Aside from randomly generating new samples, our model also facilitates applications such as retargeting, outpainting and local editing. Through extensive qualitative and quantitative evaluation, we show that our method outperforms prior methods in generation quality of 3D shapes.

Henrique Teles Maia, Chang Xiao, Dingzeyu Li et al.

Neural network applications have become popular in both enterprise and personal settings. Network solutions are tuned meticulously for each task, and designs that can robustly resolve queries end up in high demand. As the commercial value of accurate and performant machine learning models increases, so too does the demand to protect neural architectures as confidential investments. We explore the vulnerability of neural networks deployed as black boxes across accelerated hardware through electromagnetic side channels. We examine the magnetic flux emanating from a graphics processing unit's power cable, as acquired by a cheap $3 induction sensor, and find that this signal betrays the detailed topology and hyperparameters of a black-box neural network model. The attack acquires the magnetic signal for one query with unknown input values, but known input dimensions. The network reconstruction is possible due to the modular layer sequence in which deep neural networks are evaluated. We find that each layer component's evaluation produces an identifiable magnetic signal signature, from which layer topology, width, function type, and sequence order can be inferred using a suitably trained classifier and a joint consistency optimization based on integer programming. We study the extent to which network specifications can be recovered, and consider metrics for comparing network similarity. We demonstrate the potential accuracy of this side channel attack in recovering the details for a broad range of network architectures, including random designs. We consider applications that may exploit this novel side channel exposure, such as adversarial transfer attacks. In response, we discuss countermeasures to protect against our method and other similar snooping techniques.

Rundi Wu, Chang Xiao, Changxi Zheng

Deep generative models of 3D shapes have received a great deal of research interest. Yet, almost all of them generate discrete shape representations, such as voxels, point clouds, and polygon meshes. We present the first 3D generative model for a drastically different shape representation --- describing a shape as a sequence of computer-aided design (CAD) operations. Unlike meshes and point clouds, CAD models encode the user creation process of 3D shapes, widely used in numerous industrial and engineering design tasks. However, the sequential and irregular structure of CAD operations poses significant challenges for existing 3D generative models. Drawing an analogy between CAD operations and natural language, we propose a CAD generative network based on the Transformer. We demonstrate the performance of our model for both shape autoencoding and random shape generation. To train our network, we create a new CAD dataset consisting of 178,238 models and their CAD construction sequences. We have made this dataset publicly available to promote future research on this topic.

Jianjin Xu, Changxi Zheng

Generative Adversarial Networks (GANs) are able to generate high-quality images, but it remains difficult to explicitly specify the semantics of synthesized images. In this work, we aim to better understand the semantic representation of GANs, and thereby enable semantic control in GAN's generation process. Interestingly, we find that a well-trained GAN encodes image semantics in its internal feature maps in a surprisingly simple way: a linear transformation of feature maps suffices to extract the generated image semantics. To verify this simplicity, we conduct extensive experiments on various GANs and datasets; and thanks to this simplicity, we are able to learn a semantic segmentation model for a trained GAN from a small number (e.g., 8) of labeled images. Last but not least, leveraging our findings, we propose two few-shot image editing approaches, namely Semantic-Conditional Sampling and Semantic Image Editing. Given a trained GAN and as few as eight semantic annotations, the user is able to generate diverse images subject to a user-provided semantic layout, and control the synthesized image semantics. We have made the code publicly available.

Ruilin Xu, Rundi Wu, Yuko Ishiwaka et al.

We introduce a deep learning model for speech denoising, a long-standing challenge in audio analysis arising in numerous applications. Our approach is based on a key observation about human speech: there is often a short pause between each sentence or word. In a recorded speech signal, those pauses introduce a series of time periods during which only noise is present. We leverage these incidental silent intervals to learn a model for automatic speech denoising given only mono-channel audio. Detected silent intervals over time expose not just pure noise but its time-varying features, allowing the model to learn noise dynamics and suppress it from the speech signal. Experiments on multiple datasets confirm the pivotal role of silent interval detection for speech denoising, and our method outperforms several state-of-the-art denoising methods, including those that accept only audio input (like ours) and those that denoise based on audiovisual input (and hence require more information). We also show that our method enjoys excellent generalization properties, such as denoising spoken languages not seen during training.

Chang Xiao, Changxi Zheng

Modern image classification systems are often built on deep neural networks, which suffer from adversarial examples--images with deliberately crafted, imperceptible noise to mislead the network's classification. To defend against adversarial examples, a plausible idea is to obfuscate the network's gradient with respect to the input image. This general idea has inspired a long line of defense methods. Yet, almost all of them have proven vulnerable. We revisit this seemingly flawed idea from a radically different perspective. We embrace the omnipresence of adversarial examples and the numerical procedure of crafting them, and turn this harmful attacking process into a useful defense mechanism. Our defense method is conceptually simple: before feeding an input image for classification, transform it by finding an adversarial example on a pre-trained external model. We evaluate our method against a wide range of possible attacks. On both CIFAR-10 and Tiny ImageNet datasets, our method is significantly more robust than state-of-the-art methods. Particularly, in comparison to adversarial training, our method offers lower training cost as well as stronger robustness.

Chang Xiao, Peilin Zhong, Changxi Zheng

We propose a simple change to existing neural network structures for better defending against gradient-based adversarial attacks. Instead of using popular activation functions (such as ReLU), we advocate the use of k-Winners-Take-All (k-WTA) activation, a C0 discontinuous function that purposely invalidates the neural network model's gradient at densely distributed input data points. The proposed k-WTA activation can be readily used in nearly all existing networks and training methods with no significant overhead. Our proposal is theoretically rationalized. We analyze why the discontinuities in k-WTA networks can largely prevent gradient-based search of adversarial examples and why they at the same time remain innocuous to the network training. This understanding is also empirically backed. We test k-WTA activation on various network structures optimized by a training method, be it adversarial training or not. In all cases, the robustness of k-WTA networks outperforms that of traditional networks under white-box attacks.

Peilin Zhong, Yuchen Mo, Chang Xiao et al.

Many generative models have to combat $\textit{missing modes}$. The conventional wisdom to this end is by reducing through training a statistical distance (such as $f$-divergence) between the generated distribution and provided data distribution. But this is more of a heuristic than a guarantee. The statistical distance measures a $\textit{global}$, but not $\textit{local}$, similarity between two distributions. Even if it is small, it does not imply a plausible mode coverage. Rethinking this problem from a game-theoretic perspective, we show that a complete mode coverage is firmly attainable. If a generative model can approximate a data distribution moderately well under a global statistical distance measure, then we will be able to find a mixture of generators that collectively covers $\textit{every}$ data point and thus $\textit{every}$ mode, with a lower-bounded generation probability. Constructing the generator mixture has a connection to the multiplicative weights update rule, upon which we propose our algorithm. We prove that our algorithm guarantees complete mode coverage. And our experiments on real and synthetic datasets confirm better mode coverage over recent approaches, ones that also use generator mixtures but rely on global statistical distances.

Chang Xiao, Peilin Zhong, Changxi Zheng

This paper addresses the mode collapse for generative adversarial networks (GANs). We view modes as a geometric structure of data distribution in a metric space. Under this geometric lens, we embed subsamples of the dataset from an arbitrary metric space into the l2 space, while preserving their pairwise distance distribution. Not only does this metric embedding determine the dimensionality of the latent space automatically, it also enables us to construct a mixture of Gaussians to draw latent space random vectors. We use the Gaussian mixture model in tandem with a simple augmentation of the objective function to train GANs. Every major step of our method is supported by theoretical analysis, and our experiments on real and synthetic data confirm that the generator is able to produce samples spreading over most of the modes while avoiding unwanted samples, outperforming several recent GAN variants on a number of metrics and offering new features.

Dingzeyu Li, Timothy R. Langlois, Changxi Zheng

Although 360\textdegree{} cameras ease the capture of panoramic footage, it remains challenging to add realistic 360\textdegree{} audio that blends into the captured scene and is synchronized with the camera motion. We present a method for adding scene-aware spatial audio to 360\textdegree{} videos in typical indoor scenes, using only a conventional mono-channel microphone and a speaker. We observe that the late reverberation of a room's impulse response is usually diffuse spatially and directionally. Exploiting this fact, we propose a method that synthesizes the directional impulse response between any source and listening locations by combining a synthesized early reverberation part and a measured late reverberation tail. The early reverberation is simulated using a geometric acoustic simulation and then enhanced using a frequency modulation method to capture room resonances. The late reverberation is extracted from a recorded impulse response, with a carefully chosen time duration that separates out the late reverberation from the early reverberation. In our validations, we show that our synthesized spatial audio matches closely with recordings using ambisonic microphones. Lastly, we demonstrate the strength of our method in several applications.

Chang Xiao, Cheng Zhang, Changxi Zheng

We introduce FontCode, an information embedding technique for text documents. Provided a text document with specific fonts, our method embeds user-specified information in the text by perturbing the glyphs of text characters while preserving the text content. We devise an algorithm to chooses unobtrusive yet machine-recognizable glyph perturbations, leveraging a recently developed generative model that alters the glyphs of each character continuously on a font manifold. We then introduce an algorithm that embeds a user-provided message in the text document and produces an encoded document whose appearance is minimally perturbed from the original document. We also present a glyph recognition method that recovers the embedded information from an encoded document stored as a vector graphic or pixel image, or even on a printed paper. In addition, we introduce a new error-correction coding scheme that rectifies a certain number of recognition errors. Lastly, we demonstrate that our technique enables a wide array of applications, using it as a text document metadata holder, an unobtrusive optical barcode, a cryptographic message embedding scheme, and a text document signature.

Dingzeyu Li, Avinash S. Nair, Shree K. Nayar et al.

We present AirCode, a technique that allows the user to tag physically fabricated objects with given information. An AirCode tag consists of a group of carefully designed air pockets placed beneath the object surface. These air pockets are easily produced during the fabrication process of the object, without any additional material or postprocessing. Meanwhile, the air pockets affect only the scattering light transport under the surface, and thus are hard to notice to our naked eyes. But, by using a computational imaging method, the tags become detectable. We present a tool that automates the design of air pockets for the user to encode information. AirCode system also allows the user to retrieve the information from captured images via a robust decoding algorithm. We demonstrate our tagging technique with applications for metadata embedding, robotic grasping, as well as conveying object affordances.

Yun Jiang, Marcus Lim, Changxi Zheng et al.

Placing is a necessary skill for a personal robot to have in order to perform tasks such as arranging objects in a disorganized room. The object placements should not only be stable but also be in their semantically preferred placing areas and orientations. This is challenging because an environment can have a large variety of objects and placing areas that may not have been seen by the robot before. In this paper, we propose a learning approach for placing multiple objects in different placing areas in a scene. Given point-clouds of the objects and the scene, we design appropriate features and use a graphical model to encode various properties, such as the stacking of objects, stability, object-area relationship and common placing constraints. The inference in our model is an integer linear program, which we solve efficiently via an LP relaxation. We extensively evaluate our approach on 98 objects from 16 categories being placed into 40 areas. Our robotic experiments show a success rate of 98% in placing known objects and 82% in placing new objects stably. We use our method on our robots for performing tasks such as loading several dish-racks, a bookshelf and a fridge with multiple items.