Keep It Simple: Multi-Key Episodic Memory Retrieval for Ultra-Long Video Understanding

When videos extend from hours to days, directly processing them end-to-end becomes impractical for current Multi-modal Large Language Models (MLLMs). This ultra-long setting necessitates a two-stage paradigm: query-agnostic memory construction followed by retrieval-based inference. Prior work invests in complex memory construction to pre-model high-level relations in videos, despite not knowing the downstream query at build time. We instead prioritize high-recall retrievability during memory building, and defer query-specific, high-level relation composition to inference time. To this end, we propose MERIT(Multi-key Episodic Retrieval with Inference-time Temporal expansion), a simple yet effective agentic framework for ultra-long video understanding. First, we formulate an episodic multi-key representation that enables precise retrieval of fine-grained memories through a simple key-matching mechanism. Second, we introduce a neighbor filtering mechanism to capture broader semantic context without the massive computational overhead of global memory construction. This is achieved by expanding the temporal scope exclusively around the retrieved segments at inference time. By leveraging simple key-matching with this on-demand temporal expansion, MERIT achieves state-of-the-art performance across three long-video benchmarks: EgoLifeQA, LVBench, and Video-MME (Long).

Unified Video Dense Prediction from Disjoint Data

Scene understanding requires simultaneous prediction about geometry, appearance, and semantics. However, existing task-specific annotations are fragmented across incompatible, domain-specific datasets. Current unified systems circumvent this by restricting training to fully co-annotated data, or by incurring the large computational cost of pseudo-labeling. To mitigate this, we introduce UniD, a unified video model that jointly predicts eight dense scene properties-depth, surface normals, semantic segmentation, boundaries, human parts, albedo, shading, and materials-all learned from disjoint, domain-specific datasets. We propose a simple yet effective distillation step in which per-task experts supervise a unified backbone through lightweight task projectors, eliminating the need for annotation overlap or pseudo-labeling. Our key insight is that the strong visual priors of a pretrained diffusion model are sufficient to bridge the domain gaps introduced by disjoint training sources, enabling robust generalization to scene-task combinations never seen during training. UniD achieves competitive performance against per-task specialists and multi-task baselines, with strong generalization to out-of-distribution scenarios and enhanced temporal and cross-task consistency. Code and video results are available at https://unid-video.github.io/.

HarmoVid: Relightful Video Portrait Harmonization

We present a method for harmonizing the lighting of a foreground video to match a target background scene, adjusting shadows, color tone, and illumination intensity (relightful harmonization). Unlike images, acquiring labeled data for videos, where identical motions are recorded under different lighting conditions, is practically infeasible and non-scalable. While one way to create such paired data is to apply existing image-based harmonization models frame by frame to a video, the resulting outputs often suffer from significant temporal jitters. We overcome this problem by introducing a novel lighting deflickering model that can stabilize the global and local lighting flickering artifacts. Our video diffusion model learns from these upgraded deflickered data with a volume of real and synthetic videos to generate high-quality video harmonization results. We further propose an asymmetric alpha mask conditioning technique to learn the clean boundaries from real videos. Experiments demonstrate that our model achieves strong temporal coherence, naturalness, cleaner boundaries, and physically meaningful lighting behavior, while maintaining strong relighting expressiveness compared to prior image-based and video-based harmonization methods.

DAGE: Dual-Stream Architecture for Efficient and Fine-Grained Geometry Estimation

Estimating accurate, view-consistent geometry and camera poses from uncalibrated multi-view/video inputs remains challenging - especially at high spatial resolutions and over long sequences. We present DAGE, a dual-stream transformer whose main novelty is to disentangle global coherence from fine detail. A low-resolution stream operates on aggressively downsampled frames with alternating frame/global attention to build a view-consistent representation and estimate cameras efficiently, while a high-resolution stream processes the original images per-frame to preserve sharp boundaries and small structures. A lightweight adapter fuses these streams via cross-attention, injecting global context without disturbing the pretrained single-frame pathway. This design scales resolution and clip length independently, supports inputs up to 2K, and maintains practical inference cost. DAGE delivers sharp depth/pointmaps, strong cross-view consistency, and accurate poses, establishing new state-of-the-art results for video geometry estimation and multi-view reconstruction.

VideoMaMa: Mask-Guided Video Matting via Generative Prior

Generalizing video matting models to real-world videos remains a significant challenge due to the scarcity of labeled data. To address this, we present Video Mask-to-Matte Model (VideoMaMa) that converts coarse segmentation masks into pixel accurate alpha mattes, by leveraging pretrained video diffusion models. VideoMaMa demonstrates strong zero-shot generalization to real-world footage, even though it is trained solely on synthetic data. Building on this capability, we develop a scalable pseudo-labeling pipeline for large-scale video matting and construct the Matting Anything in Video (MA-V) dataset, which offers high-quality matting annotations for more than 50K real-world videos spanning diverse scenes and motions. To validate the effectiveness of this dataset, we fine-tune the SAM2 model on MA-V to obtain SAM2-Matte, which outperforms the same model trained on existing matting datasets in terms of robustness on in-the-wild videos. These findings emphasize the importance of large-scale pseudo-labeled video matting and showcase how generative priors and accessible segmentation cues can drive scalable progress in video matting research.

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.

Learning to Track Any Points from Human Motion

Human motion, with its inherent complexities, such as non-rigid deformations, articulated movements, clothing distortions, and frequent occlusions caused by limbs or other individuals, provides a rich and challenging source of supervision that is crucial for training robust and generalizable point trackers. Despite the suitability of human motion, acquiring extensive training data for point tracking remains difficult due to laborious manual annotation. Our proposed pipeline, AnthroTAP, addresses this by proposing an automated pipeline to generate pseudo-labeled training data, leveraging the Skinned Multi-Person Linear (SMPL) model. We first fit the SMPL model to detected humans in video frames, project the resulting 3D mesh vertices onto 2D image planes to generate pseudo-trajectories, handle occlusions using ray-casting, and filter out unreliable tracks based on optical flow consistency. A point tracking model trained on AnthroTAP annotated dataset achieves state-of-the-art performance on the TAP-Vid benchmark, surpassing other models trained on real videos while using 10,000 times less data and only 1 day in 4 GPUs, compared to 256 GPUs used in recent state-of-the-art.

FRAME: Pre-Training Video Feature Representations via Anticipation and Memory

Dense video prediction tasks, such as object tracking and semantic segmentation, require video encoders that generate temporally consistent, spatially dense features for every frame. However, existing approaches fall short: image encoders like DINO or CLIP lack temporal awareness, while video models such as VideoMAE underperform compared to image encoders on dense prediction tasks. We address this gap with FRAME, a self-supervised video frame encoder tailored for dense video understanding. FRAME learns to predict current and future DINO patch features from past and present RGB frames, leading to spatially precise and temporally coherent representations. To our knowledge, FRAME is the first video encoder to leverage image-based models for dense prediction while outperforming them on tasks requiring fine-grained visual correspondence. As an auxiliary capability, FRAME aligns its class token with CLIP's semantic space, supporting language-driven tasks such as video classification. We evaluate FRAME across six dense prediction tasks on seven datasets, where it consistently outperforms image encoders and existing self-supervised video models. Despite its versatility, FRAME maintains a compact architecture suitable for a range of downstream applications.

Seurat: From Moving Points to Depth

Accurate depth estimation from monocular videos remains challenging due to ambiguities inherent in single-view geometry, as crucial depth cues like stereopsis are absent. However, humans often perceive relative depth intuitively by observing variations in the size and spacing of objects as they move. Inspired by this, we propose a novel method that infers relative depth by examining the spatial relationships and temporal evolution of a set of tracked 2D trajectories. Specifically, we use off-the-shelf point tracking models to capture 2D trajectories. Then, our approach employs spatial and temporal transformers to process these trajectories and directly infer depth changes over time. Evaluated on the TAPVid-3D benchmark, our method demonstrates robust zero-shot performance, generalizing effectively from synthetic to real-world datasets. Results indicate that our approach achieves temporally smooth, high-accuracy depth predictions across diverse domains.

Tuning-Free Multi-Event Long Video Generation via Synchronized Coupled Sampling

While recent advancements in text-to-video diffusion models enable high-quality short video generation from a single prompt, generating real-world long videos in a single pass remains challenging due to limited data and high computational costs. To address this, several works propose tuning-free approaches, i.e., extending existing models for long video generation, specifically using multiple prompts to allow for dynamic and controlled content changes. However, these methods primarily focus on ensuring smooth transitions between adjacent frames, often leading to content drift and a gradual loss of semantic coherence over longer sequences. To tackle such an issue, we propose Synchronized Coupled Sampling (SynCoS), a novel inference framework that synchronizes denoising paths across the entire video, ensuring long-range consistency across both adjacent and distant frames. Our approach combines two complementary sampling strategies: reverse and optimization-based sampling, which ensure seamless local transitions and enforce global coherence, respectively. However, directly alternating between these samplings misaligns denoising trajectories, disrupting prompt guidance and introducing unintended content changes as they operate independently. To resolve this, SynCoS synchronizes them through a grounded timestep and a fixed baseline noise, ensuring fully coupled sampling with aligned denoising paths. Extensive experiments show that SynCoS significantly improves multi-event long video generation, achieving smoother transitions and superior long-range coherence, outperforming previous approaches both quantitatively and qualitatively.