Understanding and Accelerating the Training of Masked Diffusion Language Models

Masked diffusion models (MDMs) have emerged as a promising alternative to autoregressive models (ARMs) for language modeling. However, MDMs are known to learn substantially more slowly than ARMs, which may become problematic when scaling MDMs to larger models. Therefore, we ask the following question: how can we accelerate standard MDM training while maintaining its final performance? To this end, we first provide a detailed analysis of why MDM training is slow. We find that the main factor is the locality bias of language: the predictive information for a token is concentrated in nearby positions. We further investigate how this bias slows learning and suggest a simple yet effective remedy: bell-shaped time sampling as a training strategy. Notably, MDMs trained with our training recipe reach the same validation negative log-likelihood (NLL) up to $\sim4\times$ faster than standard training on One Billion Word Benchmark (LM1B). We also show faster improvements in generative perplexity, zero-shot perplexity, and downstream task performance on various benchmarks.

TrackCraft3R: Repurposing Video Diffusion Transformers for Dense 3D Tracking

Dense 3D tracking from monocular video is fundamental to dynamic scene understanding. While recent 3D foundation models provide reliable per-frame geometry, recovering object motion in this geometry remains challenging and benefits from strong motion priors learned from real-world videos. Existing 3D trackers either follow iterative paradigms trained from scratch on synthetic data or fine-tune 3D reconstruction models learned from static multi-view images, both lacking real-world motion priors. Pre-trained video diffusion transformers (video DiTs) offer rich spatio-temporal priors from internet-scale videos, making them a promising foundation for 3D tracking. However, their frame-anchored formulation, which generates each frame's content, is fundamentally mismatched with reference-anchored dense 3D tracking, which must follow the same physical points from a reference frame across time. We present TrackCraft3R, the first method to repurpose a video DiT as a feed-forward dense 3D tracker. Given a monocular video and its frame-anchored reconstruction pointmap, TrackCraft3R predicts a reference-anchored tracking pointmap that follows every pixel of the first frame across time in a single forward pass, along with its visibility. We achieve this through two designs: (i) a dual-latent representation that uses per-frame geometry latents and reference-anchored track latents as dense queries, and (ii) temporal RoPE alignment, which specifies the target timestamp of each track latent. Together, these designs convert the per-frame generative paradigm of video DiTs into a reference-anchored tracking formulation with LoRA fine-tuning. TrackCraft3R achieves state-of-the-art performance on standard sparse and dense 3D tracking benchmarks, while running 1.3x faster and using 4.6x less peak memory than the strongest prior method. We further demonstrate robustness to large motions and long videos.

Entropy-Gradient Grounding: Training-Free Evidence Retrieval in Vision-Language Models

Despite rapid progress, pretrained vision-language models still struggle when answers depend on tiny visual details or on combining clues spread across multiple regions, as in documents and compositional queries. We address this by framing grounding as test-time evidence retrieval: given a query, the model should actively identify where to look next to resolve ambiguity. To this end, we propose a training-free, model-intrinsic grounding method that uses uncertainty as supervision. Specifically, we compute the entropy of the model's next-token distribution and backpropagate it to the visual token embeddings to obtain an entropy-gradient relevance map, without auxiliary detectors or attention-map heuristics. We then extract and rank multiple coherent regions to support multi-evidence queries, and introduce an iterative zoom-and-reground procedure with a spatial-entropy stopping rule to avoid over-refinement. Experiments on seven benchmarks across four VLM architectures demonstrate consistent improvements over existing methods, with the largest gains on detail-critical and high-resolution settings, while also producing more interpretable evidence localizations.

AgentRVOS: Reasoning over Object Tracks for Zero-Shot Referring Video Object Segmentation

Referring Video Object Segmentation (RVOS) aims to segment a target object throughout a video given a natural language query. Training-free methods for this task follow a common pipeline: a MLLM selects keyframes, grounds the referred object within those frames, and a video segmentation model propagates the results. While intuitive, this design asks the MLLM to make temporal decisions before any object-level evidence is available, limiting both reasoning quality and spatio-temporal coverage. To overcome this, we propose AgentRVOS, a training-free agentic pipeline built on the complementary strengths of SAM3 and a MLLM. Given a concept derived from the query, SAM3 provides reliable perception over the full spatio-temporal extent through generated mask tracks. The MLLM then identifies the target through query-grounded reasoning over this object-level evidence, iteratively pruning guided by SAM3's temporal existence information. Extensive experiments show that AgentRVOS achieves state-of-the-art performance among training-free methods across multiple benchmarks, with consistent results across diverse MLLM backbones. Our project page is available at: https://cvlab-kaist.github.io/AgentRVOS/.

DA-Flow: Degradation-Aware Optical Flow Estimation with Diffusion Models

Optical flow models trained on high-quality data often degrade severely when confronted with real-world corruptions such as blur, noise, and compression artifacts. To overcome this limitation, we formulate Degradation-Aware Optical Flow, a new task targeting accurate dense correspondence estimation from real-world corrupted videos. Our key insight is that the intermediate representations of image restoration diffusion models are inherently corruption-aware but lack temporal awareness. To address this limitation, we lift the model to attend across adjacent frames via full spatio-temporal attention, and empirically demonstrate that the resulting features exhibit zero-shot correspondence capabilities. Based on this finding, we present DA-Flow, a hybrid architecture that fuses these diffusion features with convolutional features within an iterative refinement framework. DA-Flow substantially outperforms existing optical flow methods under severe degradation across multiple benchmarks.

TETO: Tracking Events with Teacher Observation for Motion Estimation and Frame Interpolation

Event cameras capture per-pixel brightness changes with microsecond resolution, offering continuous motion information lost between RGB frames. However, existing event-based motion estimators depend on large-scale synthetic data that often suffers from a significant sim-to-real gap. We propose TETO (Tracking Events with Teacher Observation), a teacher-student framework that learns event motion estimation from only $\sim$25 minutes of unannotated real-world recordings through knowledge distillation from a pretrained RGB tracker. Our motion-aware data curation and query sampling strategy maximizes learning from limited data by disentangling object motion from dominant ego-motion. The resulting estimator jointly predicts point trajectories and dense optical flow, which we leverage as explicit motion priors to condition a pretrained video diffusion transformer for frame interpolation. We achieve state-of-the-art point tracking on EVIMO2 and optical flow on DSEC using orders of magnitude less training data, and demonstrate that accurate motion estimation translates directly to superior frame interpolation quality on BS-ERGB and HQ-EVFI.

Repurposing Geometric Foundation Models for Multi-view Diffusion

While recent advances in generative latent spaces have driven substantial progress in single-image generation, the optimal latent space for novel view synthesis (NVS) remains largely unexplored. In particular, NVS requires geometrically consistent generation across viewpoints, but existing approaches typically operate in a view-independent VAE latent space. In this paper, we propose Geometric Latent Diffusion (GLD), a framework that repurposes the geometrically consistent feature space of geometric foundation models as the latent space for multi-view diffusion. We show that these features not only support high-fidelity RGB reconstruction but also encode strong cross-view geometric correspondences, providing a well-suited latent space for NVS. Our experiments demonstrate that GLD outperforms both VAE and RAE on 2D image quality and 3D consistency metrics, while accelerating training by more than 4.4x compared to the VAE latent space. Notably, GLD remains competitive with state-of-the-art methods that leverage large-scale text-to-image pretraining, despite training its diffusion model from scratch without such generative pretraining.

WorldCam: Interactive Autoregressive 3D Gaming Worlds with Camera Pose as a Unifying Geometric Representation

Recent advances in video diffusion transformers have enabled interactive gaming world models that allow users to explore generated environments over extended horizons. However, existing approaches struggle with precise action control and long-horizon 3D consistency. Most prior works treat user actions as abstract conditioning signals, overlooking the fundamental geometric coupling between actions and the 3D world, whereby actions induce relative camera motions that accumulate into a global camera pose within a 3D world. In this paper, we establish camera pose as a unifying geometric representation to jointly ground immediate action control and long-term 3D consistency. First, we define a physics-based continuous action space and represent user inputs in the Lie algebra to derive precise 6-DoF camera poses, which are injected into the generative model via a camera embedder to ensure accurate action alignment. Second, we use global camera poses as spatial indices to retrieve relevant past observations, enabling geometrically consistent revisiting of locations during long-horizon navigation. To support this research, we introduce a large-scale dataset comprising 3,000 minutes of authentic human gameplay annotated with camera trajectories and textual descriptions. Extensive experiments show that our approach substantially outperforms state-of-the-art interactive gaming world models in action controllability, long-horizon visual quality, and 3D spatial consistency.

Grounding World Simulation Models in a Real-World Metropolis

What if a world simulation model could render not an imagined environment but a city that actually exists? Prior generative world models synthesize visually plausible yet artificial environments by imagining all content. We present Seoul World Model (SWM), a city-scale world model grounded in the real city of Seoul. SWM anchors autoregressive video generation through retrieval-augmented conditioning on nearby street-view images. However, this design introduces several challenges, including temporal misalignment between retrieved references and the dynamic target scene, limited trajectory diversity and data sparsity from vehicle-mounted captures at sparse intervals. We address these challenges through cross-temporal pairing, a large-scale synthetic dataset enabling diverse camera trajectories, and a view interpolation pipeline that synthesizes coherent training videos from sparse street-view images. We further introduce a Virtual Lookahead Sink to stabilize long-horizon generation by continuously re-grounding each chunk to a retrieved image at a future location. We evaluate SWM against recent video world models across three cities: Seoul, Busan, and Ann Arbor. SWM outperforms existing methods in generating spatially faithful, temporally consistent, long-horizon videos grounded in actual urban environments over trajectories reaching hundreds of meters, while supporting diverse camera movements and text-prompted scenario variations.

Pri4R: Learning World Dynamics for Vision-Language-Action Models with Privileged 4D Representation

Humans learn not only how their bodies move, but also how the surrounding world responds to their actions. In contrast, while recent Vision-Language-Action (VLA) models exhibit impressive semantic understanding, they often fail to capture the spatiotemporal dynamics governing physical interaction. In this paper, we introduce Pri4R, a simple yet effective approach that endows VLA models with an implicit understanding of world dynamics by leveraging privileged 4D information during training. Specifically, Pri4R augments VLAs with a lightweight point track head that predicts 3D point tracks. By injecting VLA features into this head to jointly predict future 3D trajectories, the model learns to incorporate evolving scene geometry within its shared representation space, enabling more physically aware context for precise control. Due to its architectural simplicity, Pri4R is compatible with dominant VLA design patterns with minimal changes. During inference, we run the model using the original VLA architecture unchanged; Pri4R adds no extra inputs, outputs, or computational overhead. Across simulation and real-world evaluations, Pri4R significantly improves performance on challenging manipulation tasks, including a +10% gain on LIBERO-Long and a +40% gain on RoboCasa. We further show that 3D point track prediction is an effective supervision target for learning action-world dynamics, and validate our design choices through extensive ablations.