Abstract:Situation awareness is essential for understanding and reasoning about 3D scenes in embodied AI agents. However, existing datasets and benchmarks for situated understanding are limited in data modality, diversity, scale, and task scope. To address these limitations, we propose Multi-modal Situated Question Answering (MSQA), a large-scale multi-modal situated reasoning dataset, scalably collected leveraging 3D scene graphs and vision-language models (VLMs) across a diverse range of real-world 3D scenes. MSQA includes 251K situated question-answering pairs across 9 distinct question categories, covering complex scenarios within 3D scenes. We introduce a novel interleaved multi-modal input setting in our benchmark to provide text, image, and point cloud for situation and question description, resolving ambiguity in previous single-modality convention (e.g., text). Additionally, we devise the Multi-modal Situated Next-step Navigation (MSNN) benchmark to evaluate models' situated reasoning for navigation. Comprehensive evaluations on MSQA and MSNN highlight the limitations of existing vision-language models and underscore the importance of handling multi-modal interleaved inputs and situation modeling. Experiments on data scaling and cross-domain transfer further demonstrate the efficacy of leveraging MSQA as a pre-training dataset for developing more powerful situated reasoning models.
Abstract:Grounding natural language in physical 3D environments is essential for the advancement of embodied artificial intelligence. Current datasets and models for 3D visual grounding predominantly focus on identifying and localizing objects from static, object-centric descriptions. These approaches do not adequately address the dynamic and sequential nature of task-oriented grounding necessary for practical applications. In this work, we propose a new task: Task-oriented Sequential Grounding in 3D scenes, wherein an agent must follow detailed step-by-step instructions to complete daily activities by locating a sequence of target objects in indoor scenes. To facilitate this task, we introduce SG3D, a large-scale dataset containing 22,346 tasks with 112,236 steps across 4,895 real-world 3D scenes. The dataset is constructed using a combination of RGB-D scans from various 3D scene datasets and an automated task generation pipeline, followed by human verification for quality assurance. We adapted three state-of-the-art 3D visual grounding models to the sequential grounding task and evaluated their performance on SG3D. Our results reveal that while these models perform well on traditional benchmarks, they face significant challenges with task-oriented sequential grounding, underscoring the need for further research in this area.
Abstract:This paper presents UltraEdit, a large-scale (approximately 4 million editing samples), automatically generated dataset for instruction-based image editing. Our key idea is to address the drawbacks in existing image editing datasets like InstructPix2Pix and MagicBrush, and provide a systematic approach to producing massive and high-quality image editing samples. UltraEdit offers several distinct advantages: 1) It features a broader range of editing instructions by leveraging the creativity of large language models (LLMs) alongside in-context editing examples from human raters; 2) Its data sources are based on real images, including photographs and artworks, which provide greater diversity and reduced bias compared to datasets solely generated by text-to-image models; 3) It also supports region-based editing, enhanced by high-quality, automatically produced region annotations. Our experiments show that canonical diffusion-based editing baselines trained on UltraEdit set new records on MagicBrush and Emu-Edit benchmarks. Our analysis further confirms the crucial role of real image anchors and region-based editing data. The dataset, code, and models can be found in https://ultra-editing.github.io.
Abstract:We present OmniJARVIS, a novel Vision-Language-Action (VLA) model for open-world instruction-following agents in open-world Minecraft. Compared to prior works that either emit textual goals to separate controllers or produce the control command directly, OmniJARVIS seeks a different path to ensure both strong reasoning and efficient decision-making capabilities via unified tokenization of multimodal interaction data. First, we introduce a self-supervised approach to learn a behavior encoder that produces discretized tokens for behavior trajectories $\tau$ = {$o_0$, $a_0$, $\dots$} and an imitation learning (IL) policy decoder conditioned on these tokens. These additional behavior tokens will be augmented to the vocabulary of pretrained Multimodal Language Models (MLMs). With this encoder, we then pack long-term multimodal interactions involving task instructions, memories, thoughts, observations, textual responses, behavior trajectories, etc. into unified token sequences and model them with autoregressive transformers. Thanks to the semantically meaningful behavior tokens, the resulting VLA model, OmniJARVIS, can reason (by producing chain-of-thoughts), plan, answer questions, and act (by producing behavior tokens for the IL policy decoder). OmniJARVIS demonstrates excellent performances on a comprehensive collection of atomic, programmatic, and open-ended tasks in open-world Minecraft. Our analysis further unveils the crucial design principles in interaction data formation, unified tokenization, and its scaling potentials.
Abstract:Offline Black-Box Optimization (BBO) aims at optimizing a black-box function using the knowledge from a pre-collected offline dataset of function values and corresponding input designs. However, the high-dimensional and highly-multimodal input design space of black-box function pose inherent challenges for most existing methods that model and operate directly upon input designs. These issues include but are not limited to high sample complexity, which relates to inaccurate approximation of black-box function; and insufficient coverage and exploration of input design modes, which leads to suboptimal proposal of new input designs. In this work, we consider finding a latent space that serves as a compressed yet accurate representation of the design-value joint space, enabling effective latent exploration of high-value input design modes. To this end, we formulate an learnable energy-based latent space, and propose Noise-intensified Telescoping density-Ratio Estimation (NTRE) scheme for variational learning of an accurate latent space model without costly Markov Chain Monte Carlo. The optimization process is then exploration of high-value designs guided by the learned energy-based model in the latent space, formulated as gradient-based sampling from a latent-variable-parameterized inverse model. We show that our particular parameterization encourages expanded exploration around high-value design modes, motivated by inversion thinking of a fundamental result of conditional covariance matrix typically used for variance reduction. We observe that our method, backed by an accurately learned informative latent space and an expanding-exploration model design, yields significant improvements over strong previous methods on both synthetic and real world datasets such as the design-bench suite.
Abstract:A unified model for 3D vision-language (3D-VL) understanding is expected to take various scene representations and perform a wide range of tasks in a 3D scene. However, a considerable gap exists between existing methods and such a unified model, due to the independent application of representation and insufficient exploration of 3D multi-task training. In this paper, we introduce PQ3D, a unified model capable of using Promptable Queries to tackle a wide range of 3D-VL tasks, from low-level instance segmentation to high-level reasoning and planning. This is achieved through three key innovations: (1) unifying various 3D scene representations (i.e., voxels, point clouds, multi-view images) into a shared 3D coordinate space by segment-level grouping, (2) an attention-based query decoder for task-specific information retrieval guided by prompts, and (3) universal output heads for different tasks to support multi-task training. Tested across ten diverse 3D-VL datasets, PQ3D demonstrates impressive performance on these tasks, setting new records on most benchmarks. Particularly, PQ3D improves the state-of-the-art on ScanNet200 by 1.8% (AP), ScanRefer by 5.4% (acc@0.5), Multi3DRefer by 11.7% (F1@0.5), and Scan2Cap by 13.4% (CIDEr@0.5). Moreover, PQ3D supports flexible inference with individual or combined forms of available 3D representations, e.g., solely voxel input.
Abstract:Open-vocabulary 3D scene understanding presents a significant challenge in computer vision, withwide-ranging applications in embodied agents and augmented reality systems. Previous approaches haveadopted Neural Radiance Fields (NeRFs) to analyze 3D scenes. In this paper, we introduce SemanticGaussians, a novel open-vocabulary scene understanding approach based on 3D Gaussian Splatting. Our keyidea is distilling pre-trained 2D semantics into 3D Gaussians. We design a versatile projection approachthat maps various 2Dsemantic features from pre-trained image encoders into a novel semantic component of 3D Gaussians, withoutthe additional training required by NeRFs. We further build a 3D semantic network that directly predictsthe semantic component from raw 3D Gaussians for fast inference. We explore several applications ofSemantic Gaussians: semantic segmentation on ScanNet-20, where our approach attains a 4.2% mIoU and 4.0%mAcc improvement over prior open-vocabulary scene understanding counterparts; object part segmentation,sceneediting, and spatial-temporal segmentation with better qualitative results over 2D and 3D baselines,highlighting its versatility and effectiveness on supporting diverse downstream tasks.
Abstract:We explore how reconciling several foundation models (large language models and vision-language models) with a novel unified memory mechanism could tackle the challenging video understanding problem, especially capturing the long-term temporal relations in lengthy videos. In particular, the proposed multimodal agent VideoAgent: 1) constructs a structured memory to store both the generic temporal event descriptions and object-centric tracking states of the video; 2) given an input task query, it employs tools including video segment localization and object memory querying along with other visual foundation models to interactively solve the task, utilizing the zero-shot tool-use ability of LLMs. VideoAgent demonstrates impressive performances on several long-horizon video understanding benchmarks, an average increase of 6.6% on NExT-QA and 26.0% on EgoSchema over baselines, closing the gap between open-sourced models and private counterparts including Gemini 1.5 Pro.
Abstract:We explore how iterative revising a chain of thoughts with the help of information retrieval significantly improves large language models' reasoning and generation ability in long-horizon generation tasks, while hugely mitigating hallucination. In particular, the proposed method -- *retrieval-augmented thoughts* (RAT) -- revises each thought step one by one with retrieved information relevant to the task query, the current and the past thought steps, after the initial zero-shot CoT is generated. Applying RAT to GPT-3.5, GPT-4, and CodeLLaMA-7b substantially improves their performances on various long-horizon generation tasks; on average of relatively increasing rating scores by 13.63% on code generation, 16.96% on mathematical reasoning, 19.2% on creative writing, and 42.78% on embodied task planning. The demo page can be found at https://craftjarvis.github.io/RAT
Abstract:Leveraging large language models (LLMs) to integrate off-the-shelf tools (e.g., visual models and image processing functions) is a promising research direction to build powerful visual assistants for solving diverse visual tasks. However, the learning capability is rarely explored in existing methods, as they freeze the used tools after deployment, thereby limiting the generalization to new environments requiring specific knowledge. In this paper, we propose CLOVA, a Closed-LOop Visual Assistant to address this limitation, which encompasses inference, reflection, and learning phases in a closed-loop framework. During inference, LLMs generate programs and execute corresponding tools to accomplish given tasks. The reflection phase introduces a multimodal global-local reflection scheme to analyze whether and which tool needs to be updated based on environmental feedback. Lastly, the learning phase uses three flexible manners to collect training data in real-time and introduces a novel prompt tuning scheme to update the tools, enabling CLOVA to efficiently learn specific knowledge for new environments without human involvement. Experiments show that CLOVA outperforms tool-usage methods by 5% in visual question answering and multiple-image reasoning tasks, by 10% in knowledge tagging tasks, and by 20% in image editing tasks, highlighting the significance of the learning capability for general visual assistants.