While large multimodal models excel in broad vision-language benchmarks, they often struggle with tasks requiring precise perception of low-level visual details, such as comparing line lengths or solving simple mazes. In particular, this failure mode persists in question-answering tasks about vector graphics -- images composed purely of 2D objects and shapes. To address this challenge, we propose the Visually Descriptive Language Model (VDLM), which performs text-based reasoning about vector graphics. VDLM leverages Scalable Vector Graphics (SVG) for a more precise visual description and first uses an off-the-shelf raster-to-SVG algorithm for encoding. Since existing language models cannot understand raw SVGs in a zero-shot setting, VDLM then bridges SVG with pretrained language models through a newly introduced intermediate symbolic representation, Primal Visual Description (PVD), comprising primitive attributes (e.g., shape, position, measurement) with their corresponding predicted values. PVD is task-agnostic and represents visual primitives that are universal across all vector graphics. It can be learned with procedurally generated (SVG, PVD) pairs and also enables the direct use of LLMs for generalization to complex reasoning tasks. By casting an image to a text-based representation, we can leverage the power of language models to learn alignment from SVG to visual primitives and generalize to unseen question-answering tasks. Empirical results show that VDLM achieves stronger zero-shot performance compared to state-of-the-art LMMs, such as GPT-4V, in various low-level multimodal perception and reasoning tasks on vector graphics. We additionally present extensive analyses on VDLM's performance, demonstrating that our framework offers better interpretability due to its disentangled perception and reasoning processes. Project page: https://mikewangwzhl.github.io/VDLM/
We propose 3D Congealing, a novel problem of 3D-aware alignment for 2D images capturing semantically similar objects. Given a collection of unlabeled Internet images, our goal is to associate the shared semantic parts from the inputs and aggregate the knowledge from 2D images to a shared 3D canonical space. We introduce a general framework that tackles the task without assuming shape templates, poses, or any camera parameters. At its core is a canonical 3D representation that encapsulates geometric and semantic information. The framework optimizes for the canonical representation together with the pose for each input image, and a per-image coordinate map that warps 2D pixel coordinates to the 3D canonical frame to account for the shape matching. The optimization procedure fuses prior knowledge from a pre-trained image generative model and semantic information from input images. The former provides strong knowledge guidance for this under-constraint task, while the latter provides the necessary information to mitigate the training data bias from the pre-trained model. Our framework can be used for various tasks such as correspondence matching, pose estimation, and image editing, achieving strong results on real-world image datasets under challenging illumination conditions and on in-the-wild online image collections.
The creation of large, diverse, high-quality robot manipulation datasets is an important stepping stone on the path toward more capable and robust robotic manipulation policies. However, creating such datasets is challenging: collecting robot manipulation data in diverse environments poses logistical and safety challenges and requires substantial investments in hardware and human labour. As a result, even the most general robot manipulation policies today are mostly trained on data collected in a small number of environments with limited scene and task diversity. In this work, we introduce DROID (Distributed Robot Interaction Dataset), a diverse robot manipulation dataset with 76k demonstration trajectories or 350 hours of interaction data, collected across 564 scenes and 84 tasks by 50 data collectors in North America, Asia, and Europe over the course of 12 months. We demonstrate that training with DROID leads to policies with higher performance and improved generalization ability. We open source the full dataset, policy learning code, and a detailed guide for reproducing our robot hardware setup.
Reconstructing and simulating elastic objects from visual observations is crucial for applications in computer vision and robotics. Existing methods, such as 3D Gaussians, provide modeling for 3D appearance and geometry but lack the ability to simulate physical properties or optimize parameters for heterogeneous objects. We propose Spring-Gaus, a novel framework that integrates 3D Gaussians with physics-based simulation for reconstructing and simulating elastic objects from multi-view videos. Our method utilizes a 3D Spring-Mass model, enabling the optimization of physical parameters at the individual point level while decoupling the learning of physics and appearance. This approach achieves great sample efficiency, enhances generalization, and reduces sensitivity to the distribution of simulation particles. We evaluate Spring-Gaus on both synthetic and real-world datasets, demonstrating accurate reconstruction and simulation of elastic objects. This includes future prediction and simulation under varying initial states and environmental parameters. Project page: https://zlicheng.com/spring_gaus.
We present BEHAVIOR-1K, a comprehensive simulation benchmark for human-centered robotics. BEHAVIOR-1K includes two components, guided and motivated by the results of an extensive survey on "what do you want robots to do for you?". The first is the definition of 1,000 everyday activities, grounded in 50 scenes (houses, gardens, restaurants, offices, etc.) with more than 9,000 objects annotated with rich physical and semantic properties. The second is OMNIGIBSON, a novel simulation environment that supports these activities via realistic physics simulation and rendering of rigid bodies, deformable bodies, and liquids. Our experiments indicate that the activities in BEHAVIOR-1K are long-horizon and dependent on complex manipulation skills, both of which remain a challenge for even state-of-the-art robot learning solutions. To calibrate the simulation-to-reality gap of BEHAVIOR-1K, we provide an initial study on transferring solutions learned with a mobile manipulator in a simulated apartment to its real-world counterpart. We hope that BEHAVIOR-1K's human-grounded nature, diversity, and realism make it valuable for embodied AI and robot learning research. Project website: https://behavior.stanford.edu.
We study inferring 3D object-centric scene representations from a single image. While recent methods have shown potential in unsupervised 3D object discovery from simple synthetic images, they fail to generalize to real-world scenes with visually rich and diverse objects. This limitation stems from their object representations, which entangle objects' intrinsic attributes like shape and appearance with extrinsic, viewer-centric properties such as their 3D location. To address this bottleneck, we propose Unsupervised discovery of Object-Centric neural Fields (uOCF). uOCF focuses on learning the intrinsics of objects and models the extrinsics separately. Our approach significantly improves systematic generalization, thus enabling unsupervised learning of high-fidelity object-centric scene representations from sparse real-world images. To evaluate our approach, we collect three new datasets, including two real kitchen environments. Extensive experiments show that uOCF enables unsupervised discovery of visually rich objects from a single real image, allowing applications such as 3D object segmentation and scene manipulation. Notably, uOCF demonstrates zero-shot generalization to unseen objects from a single real image. Project page: https://red-fairy.github.io/uOCF/
Learning 3D models of all animals on the Earth requires massively scaling up existing solutions. With this ultimate goal in mind, we develop 3D-Fauna, an approach that learns a pan-category deformable 3D animal model for more than 100 animal species jointly. One crucial bottleneck of modeling animals is the limited availability of training data, which we overcome by simply learning from 2D Internet images. We show that prior category-specific attempts fail to generalize to rare species with limited training images. We address this challenge by introducing the Semantic Bank of Skinned Models (SBSM), which automatically discovers a small set of base animal shapes by combining geometric inductive priors with semantic knowledge implicitly captured by an off-the-shelf self-supervised feature extractor. To train such a model, we also contribute a new large-scale dataset of diverse animal species. At inference time, given a single image of any quadruped animal, our model reconstructs an articulated 3D mesh in a feed-forward fashion within seconds.
Urban transformations have profound societal impact on both individuals and communities at large. Accurately assessing these shifts is essential for understanding their underlying causes and ensuring sustainable urban planning. Traditional measurements often encounter constraints in spatial and temporal granularity, failing to capture real-time physical changes. While street view imagery, capturing the heartbeat of urban spaces from a pedestrian point of view, can add as a high-definition, up-to-date, and on-the-ground visual proxy of urban change. We curate the largest street view time series dataset to date, and propose an end-to-end change detection model to effectively capture physical alterations in the built environment at scale. We demonstrate the effectiveness of our proposed method by benchmark comparisons with previous literature and implementing it at the city-wide level. Our approach has the potential to supplement existing dataset and serve as a fine-grained and accurate assessment of urban change.
We introduce Neural Flow Maps, a novel simulation method bridging the emerging paradigm of implicit neural representations with fluid simulation based on the theory of flow maps, to achieve state-of-the-art simulation of inviscid fluid phenomena. We devise a novel hybrid neural field representation, Spatially Sparse Neural Fields (SSNF), which fuses small neural networks with a pyramid of overlapping, multi-resolution, and spatially sparse grids, to compactly represent long-term spatiotemporal velocity fields at high accuracy. With this neural velocity buffer in hand, we compute long-term, bidirectional flow maps and their Jacobians in a mechanistically symmetric manner, to facilitate drastic accuracy improvement over existing solutions. These long-range, bidirectional flow maps enable high advection accuracy with low dissipation, which in turn facilitates high-fidelity incompressible flow simulations that manifest intricate vortical structures. We demonstrate the efficacy of our neural fluid simulation in a variety of challenging simulation scenarios, including leapfrogging vortices, colliding vortices, vortex reconnections, as well as vortex generation from moving obstacles and density differences. Our examples show increased performance over existing methods in terms of energy conservation, visual complexity, adherence to experimental observations, and preservation of detailed vortical structures.
We introduce Ponymation, a new method for learning a generative model of articulated 3D animal motions from raw, unlabeled online videos. Unlike existing approaches for motion synthesis, our model does not require any pose annotations or parametric shape models for training, and is learned purely from a collection of raw video clips obtained from the Internet. We build upon a recent work, MagicPony, which learns articulated 3D animal shapes purely from single image collections, and extend it on two fronts. First, instead of training on static images, we augment the framework with a video training pipeline that incorporates temporal regularizations, achieving more accurate and temporally consistent reconstructions. Second, we learn a generative model of the underlying articulated 3D motion sequences via a spatio-temporal transformer VAE, simply using 2D reconstruction losses without relying on any explicit pose annotations. At inference time, given a single 2D image of a new animal instance, our model reconstructs an articulated, textured 3D mesh, and generates plausible 3D animations by sampling from the learned motion latent space.