Generating realistic human videos remains a challenging task, with the most effective methods currently relying on a human motion sequence as a control signal. Existing approaches often use existing motion extracted from other videos, which restricts applications to specific motion types and global scene matching. We propose Move-in-2D, a novel approach to generate human motion sequences conditioned on a scene image, allowing for diverse motion that adapts to different scenes. Our approach utilizes a diffusion model that accepts both a scene image and text prompt as inputs, producing a motion sequence tailored to the scene. To train this model, we collect a large-scale video dataset featuring single-human activities, annotating each video with the corresponding human motion as the target output. Experiments demonstrate that our method effectively predicts human motion that aligns with the scene image after projection. Furthermore, we show that the generated motion sequence improves human motion quality in video synthesis tasks.

Large-scale video generation models have the inherent ability to realistically model natural scenes. In this paper, we demonstrate that through a careful design of a generative video propagation framework, various video tasks can be addressed in a unified way by leveraging the generative power of such models. Specifically, our framework, Gen-Prop, encodes the original video with a selective content encoder and propagates the changes made to the first frame using an image-to-video generation model. We propose a data generation scheme to cover multiple video tasks based on instance-level video segmentation datasets. Our model is trained by incorporating a mask prediction decoder head and optimizing a region-aware loss to aid the encoder to preserve the original content while the generation model propagates the modified region. This novel design opens up new possibilities: In editing scenarios, GenProp allows substantial changes to an object’s shape; for insertion, the inserted objects can exhibit independent motion; for removal, GenProp effectively removes effects like shadows and reflections from the whole video; for tracking, GenProp is capable of tracking objects and their associated effects together. Experiment results demonstrate the leading performance of our model in various video tasks, and we further provide in-depth analyses of the proposed framework.

Text-to-video generative models have made significant strides, enabling diverse applications in entertainment, advertising, and education. However, generating RGBA video, which includes alpha channels for transparency, remains a challenge due to limited datasets and the difficulty of adapting existing models. Alpha channels are crucial for visual effects (VFX), allowing transparent elements like smoke and reflections to blend seamlessly into scenes. We introduce TransPixeler, a method to extend pretrained video models for RGBA generation while retaining the original RGB capabilities. TransPixeler leverages a diffusion transformer (DiT) architecture, incorporating alpha-specific tokens and using LoRA-based fine-tuning to jointly generate RGB and alpha channels with high consistency. By optimizing attention mechanisms, TransPixeler preserves the strengths of the original RGB model and achieves strong alignment between RGB and alpha channels despite limited training data. Our approach effectively generates diverse and consistent RGBA videos, advancing the possibilities for VFX and interactive content creation. The code is available at https://wileewang.github.io/TransPixeler/.

Efficient neural representations for dynamic video scenes are critical for applications ranging from video compression to interactive simulations. Yet, existing methods often face challenges related to high memory usage, lengthy training times, and temporal consistency. To address these issues, we introduce a novel neural video representation that combines 3D Gaussian splatting with continuous camera motion modeling. By leveraging Neural ODEs, our approach learns smooth camera trajectories while maintaining an explicit 3D scene representation through Gaussians. Additionally, we introduce a spatiotemporal hierarchical learning strategy, progressively refining spatial and temporal features to enhance reconstruction quality and accelerate convergence. This memory-efficient approach achieves high-quality rendering at impressive speeds. Experimental results show that our hierarchical learning, combined with robust camera motion modeling, captures complex dynamic scenes with strong temporal consistency, achieving state-of-the-art performance across diverse video datasets in both high- and low-motion scenarios.

Long video question answering is a challenging task that involves recognizing short-term activities and reasoning about their fine-grained relationships. State-of-the-art video Large Language Models (vLLMs) hold promise as a viable solution due to their demonstrated emergent capabilities on new tasks. However, despite being trained on millions of short seconds-long videos, vLLMs are unable to understand minutes-long videos and accurately answer questions about them. To address this limitation, we propose a lightweight and self-supervised approach, Key frame-conditioned long video-LLM (Koala), that introduces learnable spatiotemporal queries to adapt pretrained vLLMs for generalizing to longer videos. Our approach introduces two new tokenizers that condition on visual tokens computed from sparse video key frames for understanding short and long video moments. We train our proposed approach on HowTo100M and demonstrate its effectiveness on zero-shot long video understanding benchmarks, where it outperforms state-of-the-art large models by 3 - 6% in absolute accuracy across all tasks. Surprisingly, we also empirically show that our approach not only helps a pretrained vLLM to understand long videos but also improves its accuracy on short-term action recognition.

The rapid development of generative AI (GenAI) models in computer vision necessitates effective evaluation methods to ensure their quality and fairness. Existing tools primarily focus on dataset quality assurance and model explainability, leaving a significant gap in GenAI output evaluation during model development. Current practices often depend on developers’ subjective visual assessments, which may lack scalability and generalizability. This paper bridges this gap by conducting a formative study with GenAI model developers in an industrial setting. Our findings led to the development of GenLens, a visual analytic interface designed for the systematic evaluation of GenAI model outputs during the early stages of model development. GenLens offers a quantifiable approach for overviewing and annotating failure cases, customizing issue tags and classifications, and aggregating annotations from multiple users to enhance collaboration. A user study with model developers reveals that GenLens effectively enhances their workflow, evidenced by high satisfaction rates and a strong intent to integrate it into their practices. This research underscores the importance of robust early-stage evaluation tools in GenAI development, contributing to the advancement of fair and high-quality GenAI models.

Badminton is a fast-paced sport that requires a strategic combination of spatial, temporal, and technical tactics. To gain a competitive edge at high-level competitions, badminton professionals frequently analyze match videos to gain insights and develop game strategies. However, the current process for analyzing matches is time-consuming and relies heavily on manual note-taking, due to the lack of automatic data collection and appropriate visualization tools. As a result, there is a gap in effectively analyzing matches and communicating insights among badminton coaches and players. This work proposes an end-to-end immersive match analysis pipeline designed in close collaboration with badminton professionals, including Olympic and national coaches and players. We present VIRD, a VR Bird (i.e., shuttle) immersive analysis tool, that supports interactive badminton game analysis in an immersive environment based on 3D reconstructed game views of the match video. We propose a top-down analytic workflow that allows users to seamlessly move from a high-level match overview to a detailed game view of individual rallies and shots, using situated 3D visualizations and video. We collect 3D spatial and dynamic shot data and player poses with computer vision models and visualize them in VR. Through immersive visualizations, coaches can interactively analyze situated spatial data (player positions, poses, and shot trajectories) with flexible viewpoints while navigating between shots and rallies effectively with embodied interaction. We evaluated the usefulness of VIRD with Olympic and national-level coaches and players in real matches. Results show that immersive analytics supports effective badminton match analysis with reduced context-switching costs and enhances spatial understanding with a high sense of presence.

Objects make unique sounds under different perturbations, environment conditions, and poses relative to the listener. While prior works have modeled impact sounds and sound propagation in simulation, we lack a standard dataset of impact sound fields of real objects for audio-visual learning and calibration of the sim-to-real gap. We present Realimpact,a large-scale dataset of real object impact sounds recorded under controlled conditions. Re-alimpactcontains 150,000 recordings of impact sounds of 50 everyday objects with detailed annotations, including their impact locations, microphone locations, contact force profiles, material labels, and RGBD images.**The project page and dataset are available at https://samuelpclarke.com/realimpact/ We make preliminary attempts to use our dataset as a reference to current simulation methods for estimating object impact sounds that match the real world. Moreover, we demon-strate the usefulness of our dataset as a testbed for acoustic and audio-visual learning via the evaluation of two bench-mark tasks, including listener location classification and vi-sual acoustic matching.

A room’s acoustic properties are a product of the room’s geometry, the objects within the room, and their specific positions. A room’s acoustic properties can be characterized by its impulse response (RIR) between a source and listener location, or roughly inferred from recordings of natural signals present in the room. Variations in the positions of objects in a room can effect measurable changes in the room’s acoustic properties, as characterized by the RIR. Existing datasets of RIRs either do not systematically vary positions of objects in an environment, or they consist of only simulated RIRs. We present SoundCam, the largest dataset of unique RIRs from in-the-wild rooms publicly released to date. It includes 5,000 10-channel real-world measurements of room impulse responses and 2,000 10-channel recordings of music in three different rooms, including a controlled acoustic lab, an in-the-wild living room, and a conference room, with different humans in positions throughout each room. We show that these measurements can be used for interesting tasks, such as detecting and identifying humans, and tracking their positions.

We introduce a practical framework for synthesizing bubble-based water sounds that captures the rich inter-bubble coupling effects responsible for low-frequency acoustic emissions from bubble clouds. We propose coupled-bubble oscillator models with regularized singularities, and techniques to reduce the computational cost of time stepping with dense, time-varying mass matrices. Airborne acoustic emissions are estimated using finite-difference time-domain (FDTD) methods. We propose a simple, analytical surface-acceleration model, and a sample-and-hold GPU wavesolver that is simple and faster than prior CPU wavesolvers. Sound synthesis results are demonstrated using bubbly flows from incompressible, two-phase simulations, as well as procedurally generated examples using single-phase FLIP fluid animations. Our results demonstrate sound simulations with hundreds of thousands of bubbles, and perceptually significant frequency transformations with fuller low-frequency content.

Accurate modeling of personalized head-related transfer functions (HRTFs) is difficult but critical for applications requiring spatial audio. However, this remains challenging as experimental measurements require specialized equipment, numerical simulations require accurate head geometries and robust solvers, and data-driven methods are hungry for data. In this paper, we propose a new deep learning method that combines measurements and numerical simulations to take the best of three worlds. By learning the residual difference and establishing a high quality spatial basis, our method achieves consistently 2 dB to 2.5 dB lower spectral distortion (SD) compared to the state-of-the-art methods.

We propose a new modal sound synthesis method that rapidly estimates all acoustic transfer fields of a linear modal vibration model, and greatly reduces preprocessing costs. Instead of performing a separate frequency-domain Helmholtz radiation analysis for each mode, our method partitions vibration modes into chords using optimal mode conflation , then performs a single time-domain wave simulation for each chord. We then perform transfer deconflation on each chord's time-domain radiation field using a specialized QR solver, and thereby extract the frequency-domain transfer functions of each mode. The precomputed transfer functions are represented for fast far-field evaluation, e.g., using multipole expansions. In this paper, we propose to use a single scalar-valued Far-field Acoustic Transfer (FFAT) cube map. We describe a GPU-accelerated vector wavesolver that achieves high-throughput acoustic transfer computation at accuracy sufficient for sound synthesis. Our implementation, KleinPAT, can achieve hundred- to thousand-fold speedups compared to existing Helmholtz-based transfer solvers, thereby enabling large-scale generation of modal sound models for audio-visual applications.

We explore an integrated approach to sound generation that supports a wide variety of physics-based simulation models and computer-animated phenomena. Targeting high-quality offline sound synthesis, we seek to resolve animation-driven sound radiation with near-field scattering and diffraction effects. The core of our approach is a sharp-interface finite-difference time-domain (FDTD) wavesolver, with a series of supporting algorithms to handle rapidly deforming and vibrating embedded interfaces arising in physics-based animation sound. Once the solver rasterizes these interfaces, it must evaluate acceleration boundary conditions (BCs) that involve model-and phenomena-specific computations. We introduce acoustic shaders as a mechanism to abstract away these complexities, and describe a variety of implementations for computer animation: near-rigid objects with ringing and acceleration noise, deformable (finite element) models such as thin shells, bubble-based water, and virtual characters. Since time-domain wave synthesis is expensive, we only simulate pressure waves in a small region about each sound source, then estimate a far-field pressure signal. To further improve scalability beyond multi-threading, we propose a fully time-parallel sound synthesis method that is demonstrated on commodity cloud computing resources. In addition to presenting results for multiple animation phenomena (water, rigid, shells, kinematic deformers, etc.) we also propose 3D automatic dialogue replacement (3DADR) for virtual characters so that pre-recorded dialogue can include character movement, and near-field shadowing and scattering sound effects.

We present a novel method to enrich standard rigid-body impact models with a spatially varying coefficient of restitution map, or Bounce Map. Even state-of-the art methods in computer graphics assume that for a single rigid body, post- and pre-impact dynamics are related with a single global, constant, namely the coefficient of restitution. We first demonstrate that this assumption is highly inaccurate, even for simple objects. We then present a technique to efficiently and automatically generate a function which maps locations on the object's surface along with impact normals, to a scalar coefficient of restitution value. Furthermore, we propose a method for two-body restitution analysis, and, based on numerical experiments, estimate a practical model for combining one-body Bounce Map values to approximate the two-body coefficient of restitution. We show that our method not only improves accuracy, but also enables visually richer rigid-body simulations.