Multi-Granular Spatio-Temporal Token Merging for Training-Free Acceleration of Video LLMs

Video large language models (LLMs) achieve strong video understanding by leveraging a large number of spatio-temporal tokens, but suffer from quadratic computational scaling with token count. To address this, we propose a training-free spatio-temporal token merging method, named STTM. Our key insight is to exploit local spatial and temporal redundancy in video data which has been overlooked in prior work. STTM first transforms each frame into multi-granular spatial tokens using a coarse-to-fine search over a quadtree structure, then performs directed pairwise merging across the temporal dimension. This decomposed merging approach outperforms existing token reduction methods across six video QA benchmarks. Notably, STTM achieves a 2$\times$ speed-up with only a 0.5% accuracy drop under a 50% token budget, and a 3$\times$ speed-up with just a 2% drop under a 30% budget. Moreover, STTM is query-agnostic, allowing KV cache reuse across different questions for the same video. The project page is available at https://www.jshyun.me/projects/sttm.

Prototypes are Balanced Units for Efficient and Effective Partially Relevant Video Retrieval

In a retrieval system, simultaneously achieving search accuracy and efficiency is inherently challenging. This challenge is particularly pronounced in partially relevant video retrieval (PRVR), where incorporating more diverse context representations at varying temporal scales for each video enhances accuracy but increases computational and memory costs. To address this dichotomy, we propose a prototypical PRVR framework that encodes diverse contexts within a video into a fixed number of prototypes. We then introduce several strategies to enhance text association and video understanding within the prototypes, along with an orthogonal objective to ensure that the prototypes capture a diverse range of content. To keep the prototypes searchable via text queries while accurately encoding video contexts, we implement cross- and uni-modal reconstruction tasks. The cross-modal reconstruction task aligns the prototypes with textual features within a shared space, while the uni-modal reconstruction task preserves all video contexts during encoding. Additionally, we employ a video mixing technique to provide weak guidance to further align prototypes and associated textual representations. Extensive evaluations on TVR, ActivityNet-Captions, and QVHighlights validate the effectiveness of our approach without sacrificing efficiency.

CoMoGaussian: Continuous Motion-Aware Gaussian Splatting from Motion-Blurred Images

3D Gaussian Splatting (3DGS) has gained significant attention for their high-quality novel view rendering, motivating research to address real-world challenges. A critical issue is the camera motion blur caused by movement during exposure, which hinders accurate 3D scene reconstruction. In this study, we propose CoMoGaussian, a Continuous Motion-Aware Gaussian Splatting that reconstructs precise 3D scenes from motion-blurred images while maintaining real-time rendering speed. Considering the complex motion patterns inherent in real-world camera movements, we predict continuous camera trajectories using neural ordinary differential equations (ODEs). To ensure accurate modeling, we employ rigid body transformations, preserving the shape and size of the object but rely on the discrete integration of sampled frames. To better approximate the continuous nature of motion blur, we introduce a continuous motion refinement (CMR) transformation that refines rigid transformations by incorporating additional learnable parameters. By revisiting fundamental camera theory and leveraging advanced neural ODE techniques, we achieve precise modeling of continuous camera trajectories, leading to improved reconstruction accuracy. Extensive experiments demonstrate state-of-the-art performance both quantitatively and qualitatively on benchmark datasets, which include a wide range of motion blur scenarios, from moderate to extreme blur.

Humans as a Calibration Pattern: Dynamic 3D Scene Reconstruction from Unsynchronized and Uncalibrated Videos

Recent works on dynamic neural field reconstruction assume input from synchronized multi-view videos with known poses. These input constraints are often unmet in real-world setups, making the approach impractical. We demonstrate that unsynchronized videos with unknown poses can generate dynamic neural fields if the videos capture human motion. Humans are one of the most common dynamic subjects whose poses can be estimated using state-of-the-art methods. While noisy, the estimated human shape and pose parameters provide a decent initialization for the highly non-convex and under-constrained problem of training a consistent dynamic neural representation. Given the sequences of pose and shape of humans, we estimate the time offsets between videos, followed by camera pose estimations by analyzing 3D joint locations. Then, we train dynamic NeRF employing multiresolution rids while simultaneously refining both time offsets and camera poses. The setup still involves optimizing many parameters, therefore, we introduce a robust progressive learning strategy to stabilize the process. Experiments show that our approach achieves accurate spatiotemporal calibration and high-quality scene reconstruction in challenging conditions.

CoCoGaussian: Leveraging Circle of Confusion for Gaussian Splatting from Defocused Images

3D Gaussian Splatting (3DGS) has attracted significant attention for its high-quality novel view rendering, inspiring research to address real-world challenges. While conventional methods depend on sharp images for accurate scene reconstruction, real-world scenarios are often affected by defocus blur due to finite depth of field, making it essential to account for realistic 3D scene representation. In this study, we propose CoCoGaussian, a Circle of Confusion-aware Gaussian Splatting that enables precise 3D scene representation using only defocused images. CoCoGaussian addresses the challenge of defocus blur by modeling the Circle of Confusion (CoC) through a physically grounded approach based on the principles of photographic defocus. Exploiting 3D Gaussians, we compute the CoC diameter from depth and learnable aperture information, generating multiple Gaussians to precisely capture the CoC shape. Furthermore, we introduce a learnable scaling factor to enhance robustness and provide more flexibility in handling unreliable depth in scenes with reflective or refractive surfaces. Experiments on both synthetic and real-world datasets demonstrate that CoCoGaussian achieves state-of-the-art performance across multiple benchmarks.

A Simple Baseline with Single-encoder for Referring Image Segmentation

Referring image segmentation (RIS) requires dense vision-language interactions between visual pixels and textual words to segment objects based on a given description. However, commonly adapted dual-encoders in RIS, e.g., Swin transformer and BERT (uni-modal encoders) or CLIP (a multi-modal dual-encoder), lack dense multi-modal interactions during pre-training, leading to a gap with a pixel-level RIS task. To bridge this gap, existing RIS methods often rely on multi-modal fusion modules that interact two encoders, but this approach leads to high computational costs. In this paper, we present a novel RIS method with a single-encoder, i.e., BEiT-3, maximizing the potential of shared self-attention across all framework components. This enables seamless interactions of two modalities from input to final prediction, producing granularly aligned multi-modal features. Furthermore, we propose lightweight yet effective decoder modules, a Shared FPN and a Shared Mask Decoder, which contribute to the high efficiency of our model. Our simple baseline with a single encoder achieves outstanding performances on the RIS benchmark datasets while maintaining computational efficiency, compared to the most recent SoTA methods based on dual-encoders.

Classification Matters: Improving Video Action Detection with Class-Specific Attention

Video action detection (VAD) aims to detect actors and classify their actions in a video. We figure that VAD suffers more from classification rather than localization of actors. Hence, we analyze how prevailing methods form features for classification and find that they prioritize actor regions, yet often overlooking the essential contextual information necessary for accurate classification. Accordingly, we propose to reduce the bias toward actor and encourage paying attention to the context that is relevant to each action class. By assigning a class-dedicated query to each action class, our model can dynamically determine where to focus for effective classification. The proposed model demonstrates superior performance on three challenging benchmarks with significantly fewer parameters and less computation.

Regularizing Dynamic Radiance Fields with Kinematic Fields

This paper presents a novel approach for reconstructing dynamic radiance fields from monocular videos. We integrate kinematics with dynamic radiance fields, bridging the gap between the sparse nature of monocular videos and the real-world physics. Our method introduces the kinematic field, capturing motion through kinematic quantities: velocity, acceleration, and jerk. The kinematic field is jointly learned with the dynamic radiance field by minimizing the photometric loss without motion ground truth. We further augment our method with physics-driven regularizers grounded in kinematics. We propose physics-driven regularizers that ensure the physical validity of predicted kinematic quantities, including advective acceleration and jerk. Additionally, we control the motion trajectory based on rigidity equations formed with the predicted kinematic quantities. In experiments, our method outperforms the state-of-the-arts by capturing physical motion patterns within challenging real-world monocular videos.

Motion-Oriented Compositional Neural Radiance Fields for Monocular Dynamic Human Modeling

This paper introduces Motion-oriented Compositional Neural Radiance Fields (MoCo-NeRF), a framework designed to perform free-viewpoint rendering of monocular human videos via novel non-rigid motion modeling approach. In the context of dynamic clothed humans, complex cloth dynamics generate non-rigid motions that are intrinsically distinct from skeletal articulations and critically important for the rendering quality. The conventional approach models non-rigid motions as spatial (3D) deviations in addition to skeletal transformations. However, it is either time-consuming or challenging to achieve optimal quality due to its high learning complexity without a direct supervision. To target this problem, we propose a novel approach of modeling non-rigid motions as radiance residual fields to benefit from more direct color supervision in the rendering and utilize the rigid radiance fields as a prior to reduce the complexity of the learning process. Our approach utilizes a single multiresolution hash encoding (MHE) to concurrently learn the canonical T-pose representation from rigid skeletal motions and the radiance residual field for non-rigid motions. Additionally, to further improve both training efficiency and usability, we extend MoCo-NeRF to support simultaneous training of multiple subjects within a single framework, thanks to our effective design for modeling non-rigid motions. This scalability is achieved through the integration of a global MHE and learnable identity codes in addition to multiple local MHEs. We present extensive results on ZJU-MoCap and MonoCap, clearly demonstrating state-of-the-art performance in both single- and multi-subject settings. The code and model will be made publicly available at the project page: https://stevejaehyeok.github.io/publications/moco-nerf.

Masked Autoencoder for Unsupervised Video Summarization

Summarizing a video requires a diverse understanding of the video, ranging from recognizing scenes to evaluating how much each frame is essential enough to be selected as a summary. Self-supervised learning (SSL) is acknowledged for its robustness and flexibility to multiple downstream tasks, but the video SSL has not shown its value for dense understanding tasks like video summarization. We claim an unsupervised autoencoder with sufficient self-supervised learning does not need any extra downstream architecture design or fine-tuning weights to be utilized as a video summarization model. The proposed method to evaluate the importance score of each frame takes advantage of the reconstruction score of the autoencoder's decoder. We evaluate the method in major unsupervised video summarization benchmarks to show its effectiveness under various experimental settings.