Struct2D: A Perception-Guided Framework for Spatial Reasoning in Large Multimodal Models

Unlocking spatial reasoning in Large Multimodal Models (LMMs) is crucial for enabling intelligent interaction with 3D environments. While prior efforts often rely on explicit 3D inputs or specialized model architectures, we ask: can LMMs reason about 3D space using only structured 2D representations derived from perception? We introduce Struct2D, a perception-guided prompting framework that combines bird's-eye-view (BEV) images with object marks and object-centric metadata, optionally incorporating egocentric keyframes when needed. Using Struct2D, we conduct an in-depth zero-shot analysis of closed-source LMMs (e.g., GPT-o3) and find that they exhibit surprisingly strong spatial reasoning abilities when provided with structured 2D inputs, effectively handling tasks such as relative direction estimation and route planning. Building on these insights, we construct Struct2D-Set, a large-scale instruction tuning dataset with 200K fine-grained QA pairs across eight spatial reasoning categories, generated automatically from 3D indoor scenes. We fine-tune an open-source LMM (Qwen2.5VL) on Struct2D-Set, achieving competitive performance on multiple benchmarks, including 3D question answering, dense captioning, and object grounding. Our approach demonstrates that structured 2D inputs can effectively bridge perception and language reasoning in LMMs-without requiring explicit 3D representations as input. We will release both our code and dataset to support future research.

Object Concepts Emerge from Motion

Object concepts play a foundational role in human visual cognition, enabling perception, memory, and interaction in the physical world. Inspired by findings in developmental neuroscience - where infants are shown to acquire object understanding through observation of motion - we propose a biologically inspired framework for learning object-centric visual representations in an unsupervised manner. Our key insight is that motion boundary serves as a strong signal for object-level grouping, which can be used to derive pseudo instance supervision from raw videos. Concretely, we generate motion-based instance masks using off-the-shelf optical flow and clustering algorithms, and use them to train visual encoders via contrastive learning. Our framework is fully label-free and does not rely on camera calibration, making it scalable to large-scale unstructured video data. We evaluate our approach on three downstream tasks spanning both low-level (monocular depth estimation) and high-level (3D object detection and occupancy prediction) vision. Our models outperform previous supervised and self-supervised baselines and demonstrate strong generalization to unseen scenes. These results suggest that motion-induced object representations offer a compelling alternative to existing vision foundation models, capturing a crucial but overlooked level of abstraction: the visual instance. The corresponding code will be released upon paper acceptance.

GeoMan: Temporally Consistent Human Geometry Estimation using Image-to-Video Diffusion

Estimating accurate and temporally consistent 3D human geometry from videos is a challenging problem in computer vision. Existing methods, primarily optimized for single images, often suffer from temporal inconsistencies and fail to capture fine-grained dynamic details. To address these limitations, we present GeoMan, a novel architecture designed to produce accurate and temporally consistent depth and normal estimations from monocular human videos. GeoMan addresses two key challenges: the scarcity of high-quality 4D training data and the need for metric depth estimation to accurately model human size. To overcome the first challenge, GeoMan employs an image-based model to estimate depth and normals for the first frame of a video, which then conditions a video diffusion model, reframing video geometry estimation task as an image-to-video generation problem. This design offloads the heavy lifting of geometric estimation to the image model and simplifies the video model's role to focus on intricate details while using priors learned from large-scale video datasets. Consequently, GeoMan improves temporal consistency and generalizability while requiring minimal 4D training data. To address the challenge of accurate human size estimation, we introduce a root-relative depth representation that retains critical human-scale details and is easier to be estimated from monocular inputs, overcoming the limitations of traditional affine-invariant and metric depth representations. GeoMan achieves state-of-the-art performance in both qualitative and quantitative evaluations, demonstrating its effectiveness in overcoming longstanding challenges in 3D human geometry estimation from videos.

Occlusion Boundary and Depth: Mutual Enhancement via Multi-Task Learning

Occlusion Boundary Estimation (OBE) identifies boundaries arising from both inter-object occlusions and self-occlusion within individual objects, distinguishing intrinsic object edges from occlusion-induced contours to improve scene understanding and 3D reconstruction capacity. This is closely related to Monocular Depth Estimation (MDE), which infers depth from a single image, as occlusion boundaries provide critical geometric cues for resolving depth ambiguities, while depth priors can conversely refine occlusion reasoning in complex scenes. In this paper, we propose a novel network, MoDOT, that first jointly estimates depth and OBs. We propose CASM, a cross-attention multi-scale strip convolution module, leverages mid-level OB features to significantly enhance depth prediction. Additionally, we introduce an occlusion-aware loss function, OBDCL, which encourages sharper and more accurate depth boundaries. Extensive experiments on both real and synthetic datasets demonstrate the mutual benefits of jointly estimating depth and OB, and highlight the effectiveness of our model design. Our method achieves the state-of-the-art (SOTA) on both our proposed synthetic datasets and one popular real dataset, NYUD-v2, significantly outperforming multi-task baselines. Besides, without domain adaptation, results on real-world depth transfer are comparable to the competitors, while preserving sharp occlusion boundaries for geometric fidelity. We will release our code, pre-trained models, and datasets to support future research in this direction.

From Single Images to Motion Policies via Video-Generation Environment Representations

Autonomous robots typically need to construct representations of their surroundings and adapt their motions to the geometry of their environment. Here, we tackle the problem of constructing a policy model for collision-free motion generation, consistent with the environment, from a single input RGB image. Extracting 3D structures from a single image often involves monocular depth estimation. Developments in depth estimation have given rise to large pre-trained models such as DepthAnything. However, using outputs of these models for downstream motion generation is challenging due to frustum-shaped errors that arise. Instead, we propose a framework known as Video-Generation Environment Representation (VGER), which leverages the advances of large-scale video generation models to generate a moving camera video conditioned on the input image. Frames of this video, which form a multiview dataset, are then input into a pre-trained 3D foundation model to produce a dense point cloud. We then introduce a multi-scale noise approach to train an implicit representation of the environment structure and build a motion generation model that complies with the geometry of the representation. We extensively evaluate VGER over a diverse set of indoor and outdoor environments. We demonstrate its ability to produce smooth motions that account for the captured geometry of a scene, all from a single RGB input image.

Spatial RoboGrasp: Generalized Robotic Grasping Control Policy

Achieving generalizable and precise robotic manipulation across diverse environments remains a critical challenge, largely due to limitations in spatial perception. While prior imitation-learning approaches have made progress, their reliance on raw RGB inputs and handcrafted features often leads to overfitting and poor 3D reasoning under varied lighting, occlusion, and object conditions. In this paper, we propose a unified framework that couples robust multimodal perception with reliable grasp prediction. Our architecture fuses domain-randomized augmentation, monocular depth estimation, and a depth-aware 6-DoF Grasp Prompt into a single spatial representation for downstream action planning. Conditioned on this encoding and a high-level task prompt, our diffusion-based policy yields precise action sequences, achieving up to 40% improvement in grasp success and 45% higher task success rates under environmental variation. These results demonstrate that spatially grounded perception, paired with diffusion-based imitation learning, offers a scalable and robust solution for general-purpose robotic grasping.

4DTAM: Non-Rigid Tracking and Mapping via Dynamic Surface Gaussians

We propose the first 4D tracking and mapping method that jointly performs camera localization and non-rigid surface reconstruction via differentiable rendering. Our approach captures 4D scenes from an online stream of color images with depth measurements or predictions by jointly optimizing scene geometry, appearance, dynamics, and camera ego-motion. Although natural environments exhibit complex non-rigid motions, 4D-SLAM remains relatively underexplored due to its inherent challenges; even with 2.5D signals, the problem is ill-posed because of the high dimensionality of the optimization space. To overcome these challenges, we first introduce a SLAM method based on Gaussian surface primitives that leverages depth signals more effectively than 3D Gaussians, thereby achieving accurate surface reconstruction. To further model non-rigid deformations, we employ a warp-field represented by a multi-layer perceptron (MLP) and introduce a novel camera pose estimation technique along with surface regularization terms that facilitate spatio-temporal reconstruction. In addition to these algorithmic challenges, a significant hurdle in 4D SLAM research is the lack of reliable ground truth and evaluation protocols, primarily due to the difficulty of 4D capture using commodity sensors. To address this, we present a novel open synthetic dataset of everyday objects with diverse motions, leveraging large-scale object models and animation modeling. In summary, we open up the modern 4D-SLAM research by introducing a novel method and evaluation protocols grounded in modern vision and rendering techniques.

Joint Depth and Reflectivity Estimation using Single-Photon LiDAR

Single-Photon Light Detection and Ranging (SP-LiDAR is emerging as a leading technology for long-range, high-precision 3D vision tasks. In SP-LiDAR, timestamps encode two complementary pieces of information: pulse travel time (depth) and the number of photons reflected by the object (reflectivity). Existing SP-LiDAR reconstruction methods typically recover depth and reflectivity separately or sequentially use one modality to estimate the other. Moreover, the conventional 3D histogram construction is effective mainly for slow-moving or stationary scenes. In dynamic scenes, however, it is more efficient and effective to directly process the timestamps. In this paper, we introduce an estimation method to simultaneously recover both depth and reflectivity in fast-moving scenes. We offer two contributions: (1) A theoretical analysis demonstrating the mutual correlation between depth and reflectivity and the conditions under which joint estimation becomes beneficial. (2) A novel reconstruction method, "SPLiDER", which exploits the shared information to enhance signal recovery. On both synthetic and real SP-LiDAR data, our method outperforms existing approaches, achieving superior joint reconstruction quality.

Learning Fine-Grained Geometry for Sparse-View Splatting via Cascade Depth Loss

Novel view synthesis is a fundamental task in 3D computer vision that aims to reconstruct realistic images from a set of posed input views. However, reconstruction quality degrades significantly under sparse-view conditions due to limited geometric cues. Existing methods, such as Neural Radiance Fields (NeRF) and the more recent 3D Gaussian Splatting (3DGS), often suffer from blurred details and structural artifacts when trained with insufficient views. Recent works have identified the quality of rendered depth as a key factor in mitigating these artifacts, as it directly affects geometric accuracy and view consistency. In this paper, we address these challenges by introducing Hierarchical Depth-Guided Splatting (HDGS), a depth supervision framework that progressively refines geometry from coarse to fine levels. Central to HDGS is a novel Cascade Pearson Correlation Loss (CPCL), which aligns rendered and estimated monocular depths across multiple spatial scales. By enforcing multi-scale depth consistency, our method substantially improves structural fidelity in sparse-view scenarios. Extensive experiments on the LLFF and DTU benchmarks demonstrate that HDGS achieves state-of-the-art performance under sparse-view settings while maintaining efficient and high-quality rendering

A Shape-Aware Total Body Photography System for In-focus Surface Coverage Optimization

Total Body Photography (TBP) is becoming a useful screening tool for patients at high risk for skin cancer. While much progress has been made, existing TBP systems can be further improved for automatic detection and analysis of suspicious skin lesions, which is in part related to the resolution and sharpness of acquired images. This paper proposes a novel shape-aware TBP system automatically capturing full-body images while optimizing image quality in terms of resolution and sharpness over the body surface. The system uses depth and RGB cameras mounted on a 360-degree rotary beam, along with 3D body shape estimation and an in-focus surface optimization method to select the optimal focus distance for each camera pose. This allows for optimizing the focused coverage over the complex 3D geometry of the human body given the calibrated camera poses. We evaluate the effectiveness of the system in capturing high-fidelity body images. The proposed system achieves an average resolution of 0.068 mm/pixel and 0.0566 mm/pixel with approximately 85% and 95% of surface area in-focus, evaluated on simulation data of diverse body shapes and poses as well as a real scan of a mannequin respectively. Furthermore, the proposed shape-aware focus method outperforms existing focus protocols (e.g. auto-focus). We believe the high-fidelity imaging enabled by the proposed system will improve automated skin lesion analysis for skin cancer screening.