From Generation to Generalization: Emergent Few-Shot Learning in Video Diffusion Models

Video Diffusion Models (VDMs) have emerged as powerful generative tools, capable of synthesizing high-quality spatiotemporal content. Yet, their potential goes far beyond mere video generation. We argue that the training dynamics of VDMs, driven by the need to model coherent sequences, naturally pushes them to internalize structured representations and an implicit understanding of the visual world. To probe the extent of this internal knowledge, we introduce a few-shot fine-tuning framework that repurposes VDMs for new tasks using only a handful of examples. Our method transforms each task into a visual transition, enabling the training of LoRA weights on short input-output sequences without altering the generative interface of a frozen VDM. Despite minimal supervision, the model exhibits strong generalization across diverse tasks, from low-level vision (for example, segmentation and pose estimation) to high-level reasoning (for example, on ARC-AGI). These results reframe VDMs as more than generative engines. They are adaptable visual learners with the potential to serve as the backbone for future foundation models in vision.

VoxDet: Rethinking 3D Semantic Occupancy Prediction as Dense Object Detection

3D semantic occupancy prediction aims to reconstruct the 3D geometry and semantics of the surrounding environment. With dense voxel labels, prior works typically formulate it as a dense segmentation task, independently classifying each voxel. However, this paradigm neglects critical instance-centric discriminability, leading to instance-level incompleteness and adjacent ambiguities. To address this, we highlight a free lunch of occupancy labels: the voxel-level class label implicitly provides insight at the instance level, which is overlooked by the community. Motivated by this observation, we first introduce a training-free Voxel-to-Instance (VoxNT) trick: a simple yet effective method that freely converts voxel-level class labels into instance-level offset labels. Building on this, we further propose VoxDet, an instance-centric framework that reformulates the voxel-level occupancy prediction as dense object detection by decoupling it into two sub-tasks: offset regression and semantic prediction. Specifically, based on the lifted 3D volume, VoxDet first uses (a) Spatially-decoupled Voxel Encoder to generate disentangled feature volumes for the two sub-tasks, which learn task-specific spatial deformation in the densely projected tri-perceptive space. Then, we deploy (b) Task-decoupled Dense Predictor to address this task via dense detection. Here, we first regress a 4D offset field to estimate distances (6 directions) between voxels and object borders in the voxel space. The regressed offsets are then used to guide the instance-level aggregation in the classification branch, achieving instance-aware prediction. Experiments show that VoxDet can be deployed on both camera and LiDAR input, jointly achieving state-of-the-art results on both benchmarks. VoxDet is not only highly efficient, but also achieves 63.0 IoU on the SemanticKITTI test set, ranking 1st on the online leaderboard.

X-GRM: Large Gaussian Reconstruction Model for Sparse-view X-rays to Computed Tomography

Computed Tomography serves as an indispensable tool in clinical workflows, providing non-invasive visualization of internal anatomical structures. Existing CT reconstruction works are limited to small-capacity model architecture, inflexible volume representation, and small-scale training data. In this paper, we present X-GRM (X-ray Gaussian Reconstruction Model), a large feedforward model for reconstructing 3D CT from sparse-view 2D X-ray projections. X-GRM employs a scalable transformer-based architecture to encode an arbitrary number of sparse X-ray inputs, where tokens from different views are integrated efficiently. Then, tokens are decoded into a new volume representation, named Voxel-based Gaussian Splatting (VoxGS), which enables efficient CT volume extraction and differentiable X-ray rendering. To support the training of X-GRM, we collect ReconX-15K, a large-scale CT reconstruction dataset containing around 15,000 CT/X-ray pairs across diverse organs, including the chest, abdomen, pelvis, and tooth etc. This combination of a high-capacity model, flexible volume representation, and large-scale training data empowers our model to produce high-quality reconstructions from various testing inputs, including in-domain and out-domain X-ray projections. Project Page: https://github.com/CUHK-AIM-Group/X-GRM.

CAMELTrack: Context-Aware Multi-cue ExpLoitation for Online Multi-Object Tracking

Online multi-object tracking has been recently dominated by tracking-by-detection (TbD) methods, where recent advances rely on increasingly sophisticated heuristics for tracklet representation, feature fusion, and multi-stage matching. The key strength of TbD lies in its modular design, enabling the integration of specialized off-the-shelf models like motion predictors and re-identification. However, the extensive usage of human-crafted rules for temporal associations makes these methods inherently limited in their ability to capture the complex interplay between various tracking cues. In this work, we introduce CAMEL, a novel association module for Context-Aware Multi-Cue ExpLoitation, that learns resilient association strategies directly from data, breaking free from hand-crafted heuristics while maintaining TbD's valuable modularity. At its core, CAMEL employs two transformer-based modules and relies on a novel association-centric training scheme to effectively model the complex interactions between tracked targets and their various association cues. Unlike end-to-end detection-by-tracking approaches, our method remains lightweight and fast to train while being able to leverage external off-the-shelf models. Our proposed online tracking pipeline, CAMELTrack, achieves state-of-the-art performance on multiple tracking benchmarks. Our code is available at https://github.com/TrackingLaboratory/CAMELTrack.

Weak-for-Strong: Training Weak Meta-Agent to Harness Strong Executors

Efficiently leveraging of the capabilities of contemporary large language models (LLMs) is increasingly challenging, particularly when direct fine-tuning is expensive and often impractical. Existing training-free methods, including manually or automated designed workflows, typically demand substantial human effort or yield suboptimal results. This paper proposes Weak-for-Strong Harnessing (W4S), a novel framework that customizes smaller, cost-efficient language models to design and optimize workflows for harnessing stronger models. W4S formulates workflow design as a multi-turn markov decision process and introduces reinforcement learning for agentic workflow optimization (RLAO) to train a weak meta-agent. Through iterative interaction with the environment, the meta-agent learns to design increasingly effective workflows without manual intervention. Empirical results demonstrate the superiority of W4S that our 7B meta-agent, trained with just one GPU hour, outperforms the strongest baseline by 2.9% ~ 24.6% across eleven benchmarks, successfully elevating the performance of state-of-the-art models such as GPT-3.5-Turbo and GPT-4o. Notably, W4S exhibits strong generalization capabilities across both seen and unseen tasks, offering an efficient, high-performing alternative to directly fine-tuning strong models.

Boosting Omnidirectional Stereo Matching with a Pre-trained Depth Foundation Model

Omnidirectional depth perception is essential for mobile robotics applications that require scene understanding across a full 360{\deg} field of view. Camera-based setups offer a cost-effective option by using stereo depth estimation to generate dense, high-resolution depth maps without relying on expensive active sensing. However, existing omnidirectional stereo matching approaches achieve only limited depth accuracy across diverse environments, depth ranges, and lighting conditions, due to the scarcity of real-world data. We present DFI-OmniStereo, a novel omnidirectional stereo matching method that leverages a large-scale pre-trained foundation model for relative monocular depth estimation within an iterative optimization-based stereo matching architecture. We introduce a dedicated two-stage training strategy to utilize the relative monocular depth features for our omnidirectional stereo matching before scale-invariant fine-tuning. DFI-OmniStereo achieves state-of-the-art results on the real-world Helvipad dataset, reducing disparity MAE by approximately 16% compared to the previous best omnidirectional stereo method.

FG$^2$: Fine-Grained Cross-View Localization by Fine-Grained Feature Matching

We propose a novel fine-grained cross-view localization method that estimates the 3 Degrees of Freedom pose of a ground-level image in an aerial image of the surroundings by matching fine-grained features between the two images. The pose is estimated by aligning a point plane generated from the ground image with a point plane sampled from the aerial image. To generate the ground points, we first map ground image features to a 3D point cloud. Our method then learns to select features along the height dimension to pool the 3D points to a Bird's-Eye-View (BEV) plane. This selection enables us to trace which feature in the ground image contributes to the BEV representation. Next, we sample a set of sparse matches from computed point correspondences between the two point planes and compute their relative pose using Procrustes alignment. Compared to the previous state-of-the-art, our method reduces the mean localization error by 28% on the VIGOR cross-area test set. Qualitative results show that our method learns semantically consistent matches across ground and aerial views through weakly supervised learning from the camera pose.

Unified Human Localization and Trajectory Prediction with Monocular Vision

Conventional human trajectory prediction models rely on clean curated data, requiring specialized equipment or manual labeling, which is often impractical for robotic applications. The existing predictors tend to overfit to clean observation affecting their robustness when used with noisy inputs. In this work, we propose MonoTransmotion (MT), a Transformer-based framework that uses only a monocular camera to jointly solve localization and prediction tasks. Our framework has two main modules: Bird's Eye View (BEV) localization and trajectory prediction. The BEV localization module estimates the position of a person using 2D human poses, enhanced by a novel directional loss for smoother sequential localizations. The trajectory prediction module predicts future motion from these estimates. We show that by jointly training both tasks with our unified framework, our method is more robust in real-world scenarios made of noisy inputs. We validate our MT network on both curated and non-curated datasets. On the curated dataset, MT achieves around 12% improvement over baseline models on BEV localization and trajectory prediction. On real-world non-curated dataset, experimental results indicate that MT maintains similar performance levels, highlighting its robustness and generalization capability. The code is available at https://github.com/vita-epfl/MonoTransmotion.

COARSE: Collaborative Pseudo-Labeling with Coarse Real Labels for Off-Road Semantic Segmentation

Autonomous off-road navigation faces challenges due to diverse, unstructured environments, requiring robust perception with both geometric and semantic understanding. However, scarce densely labeled semantic data limits generalization across domains. Simulated data helps, but introduces domain adaptation issues. We propose COARSE, a semi-supervised domain adaptation framework for off-road semantic segmentation, leveraging sparse, coarse in-domain labels and densely labeled out-of-domain data. Using pretrained vision transformers, we bridge domain gaps with complementary pixel-level and patch-level decoders, enhanced by a collaborative pseudo-labeling strategy on unlabeled data. Evaluations on RUGD and Rellis-3D datasets show significant improvements of 9.7\% and 8.4\% respectively, versus only using coarse data. Tests on real-world off-road vehicle data in a multi-biome setting further demonstrate COARSE's applicability.

Towards Generalizable Trajectory Prediction Using Dual-Level Representation Learning And Adaptive Prompting

Existing vehicle trajectory prediction models struggle with generalizability, prediction uncertainties, and handling complex interactions. It is often due to limitations like complex architectures customized for a specific dataset and inefficient multimodal handling. We propose Perceiver with Register queries (PerReg+), a novel trajectory prediction framework that introduces: (1) Dual-Level Representation Learning via Self-Distillation (SD) and Masked Reconstruction (MR), capturing global context and fine-grained details. Additionally, our approach of reconstructing segmentlevel trajectories and lane segments from masked inputs with query drop, enables effective use of contextual information and improves generalization; (2) Enhanced Multimodality using register-based queries and pretraining, eliminating the need for clustering and suppression; and (3) Adaptive Prompt Tuning during fine-tuning, freezing the main architecture and optimizing a small number of prompts for efficient adaptation. PerReg+ sets a new state-of-the-art performance on nuScenes [1], Argoverse 2 [2], and Waymo Open Motion Dataset (WOMD) [3]. Remarkable, our pretrained model reduces the error by 6.8% on smaller datasets, and multi-dataset training enhances generalization. In cross-domain tests, PerReg+ reduces B-FDE by 11.8% compared to its non-pretrained variant.