Zhiwen Fan

Hanxue Liang, Zhiwen Fan, Rishov Sarkar et al.

Multi-task learning (MTL) encapsulates multiple learned tasks in a single model and often lets those tasks learn better jointly. However, when deploying MTL onto those real-world systems that are often resource-constrained or latency-sensitive, two prominent challenges arise: (i) during training, simultaneously optimizing all tasks is often difficult due to gradient conflicts across tasks; (ii) at inference, current MTL regimes have to activate nearly the entire model even to just execute a single task. Yet most real systems demand only one or two tasks at each moment, and switch between tasks as needed: therefore such all tasks activated inference is also highly inefficient and non-scalable. In this paper, we present a model-accelerator co-design framework to enable efficient on-device MTL. Our framework, dubbed M$^3$ViT, customizes mixture-of-experts (MoE) layers into a vision transformer (ViT) backbone for MTL, and sparsely activates task-specific experts during training. Then at inference with any task of interest, the same design allows for activating only the task-corresponding sparse expert pathway, instead of the full model. Our new model design is further enhanced by hardware-level innovations, in particular, a novel computation reordering scheme tailored for memory-constrained MTL that achieves zero-overhead switching between tasks and can scale to any number of experts. When executing single-task inference, M$^{3}$ViT achieves higher accuracies than encoder-focused MTL methods, while significantly reducing 88% inference FLOPs. When implemented on a hardware platform of one Xilinx ZCU104 FPGA, our co-design framework reduces the memory requirement by 2.4 times, while achieving energy efficiency up to 9.23 times higher than a comparable FPGA baseline. Code is available at: https://github.com/VITA-Group/M3ViT.

Tianlong Chen, Peihao Wang, Zhiwen Fan et al.

Neural Radiance Field (NeRF) regresses a neural parameterized scene by differentially rendering multi-view images with ground-truth supervision. However, when interpolating novel views, NeRF often yields inconsistent and visually non-smooth geometric results, which we consider as a generalization gap between seen and unseen views. Recent advances in convolutional neural networks have demonstrated the promise of advanced robust data augmentations, either random or learned, in enhancing both in-distribution and out-of-distribution generalization. Inspired by that, we propose Augmented NeRF (Aug-NeRF), which for the first time brings the power of robust data augmentations into regularizing the NeRF training. Particularly, our proposal learns to seamlessly blend worst-case perturbations into three distinct levels of the NeRF pipeline with physical grounds, including (1) the input coordinates, to simulate imprecise camera parameters at image capture; (2) intermediate features, to smoothen the intrinsic feature manifold; and (3) pre-rendering output, to account for the potential degradation factors in the multi-view image supervision. Extensive results demonstrate that Aug-NeRF effectively boosts NeRF performance in both novel view synthesis (up to 1.5dB PSNR gain) and underlying geometry reconstruction. Furthermore, thanks to the implicit smooth prior injected by the triple-level augmentations, Aug-NeRF can even recover scenes from heavily corrupted images, a highly challenging setting untackled before. Our codes are available in https://github.com/VITA-Group/Aug-NeRF.

Dejia Xu, Yifan Jiang, Peihao Wang et al. · gatech

Despite the rapid development of Neural Radiance Field (NeRF), the necessity of dense covers largely prohibits its wider applications. While several recent works have attempted to address this issue, they either operate with sparse views (yet still, a few of them) or on simple objects/scenes. In this work, we consider a more ambitious task: training neural radiance field, over realistically complex visual scenes, by "looking only once", i.e., using only a single view. To attain this goal, we present a Single View NeRF (SinNeRF) framework consisting of thoughtfully designed semantic and geometry regularizations. Specifically, SinNeRF constructs a semi-supervised learning process, where we introduce and propagate geometry pseudo labels and semantic pseudo labels to guide the progressive training process. Extensive experiments are conducted on complex scene benchmarks, including NeRF synthetic dataset, Local Light Field Fusion dataset, and DTU dataset. We show that even without pre-training on multi-view datasets, SinNeRF can yield photo-realistic novel-view synthesis results. Under the single image setting, SinNeRF significantly outperforms the current state-of-the-art NeRF baselines in all cases. Project page: https://vita-group.github.io/SinNeRF/

Peihao Wang, Zhiwen Fan, Tianlong Chen et al.

Representing visual signals by coordinate-based deep fully-connected networks has been shown advantageous in fitting complex details and solving inverse problems than discrete grid-based representation. However, acquiring such a continuous Implicit Neural Representation (INR) requires tedious per-scene training on tons of signal measurements, which limits its practicality. In this paper, we present a generic INR framework that achieves both data and training efficiency by learning a Neural Implicit Dictionary (NID) from a data collection and representing INR as a functional combination of basis sampled from the dictionary. Our NID assembles a group of coordinate-based subnetworks which are tuned to span the desired function space. After training, one can instantly and robustly acquire an unseen scene representation by solving the coding coefficients. To parallelly optimize a large group of networks, we borrow the idea from Mixture-of-Expert (MoE) to design and train our network with a sparse gating mechanism. Our experiments show that, NID can improve reconstruction of 2D images or 3D scenes by 2 orders of magnitude faster with up to 98% less input data. We further demonstrate various applications of NID in image inpainting and occlusion removal, which are considered to be challenging with vanilla INR. Our codes are available in https://github.com/VITA-Group/Neural-Implicit-Dict.

Stefan Abi-Karam, Rishov Sarkar, Dejia Xu et al. · gatech

An increasing number of researchers are finding use for nth-order gradient computations for a wide variety of applications, including graphics, meta-learning (MAML), scientific computing, and most recently, implicit neural representations (INRs). Recent work shows that the gradient of an INR can be used to edit the data it represents directly without needing to convert it back to a discrete representation. However, given a function represented as a computation graph, traditional architectures face challenges in efficiently computing its nth-order gradient due to the higher demand for computing power and higher complexity in data movement. This makes it a promising target for FPGA acceleration. In this work, we introduce INR-Arch, a framework that transforms the computation graph of an nth-order gradient into a hardware-optimized dataflow architecture. We address this problem in two phases. First, we design a dataflow architecture that uses FIFO streams and an optimized computation kernel library, ensuring high memory efficiency and parallel computation. Second, we propose a compiler that extracts and optimizes computation graphs, automatically configures hardware parameters such as latency and stream depths to optimize throughput, while ensuring deadlock-free operation, and outputs High-Level Synthesis (HLS) code for FPGA implementation. We utilize INR editing as our benchmark, presenting results that demonstrate 1.8-4.8x and 1.5-3.6x speedup compared to CPU and GPU baselines respectively. Furthermore, we obtain 3.1-8.9x and 1.7-4.3x lower memory usage, and 1.7-11.3x and 5.5-32.8x lower energy-delay product. Our framework will be made open-source and available on GitHub.

Wenyan Cong, Hanxue Liang, Peihao Wang et al.

Cross-scene generalizable NeRF models, which can directly synthesize novel views of unseen scenes, have become a new spotlight of the NeRF field. Several existing attempts rely on increasingly end-to-end "neuralized" architectures, i.e., replacing scene representation and/or rendering modules with performant neural networks such as transformers, and turning novel view synthesis into a feed-forward inference pipeline. While those feedforward "neuralized" architectures still do not fit diverse scenes well out of the box, we propose to bridge them with the powerful Mixture-of-Experts (MoE) idea from large language models (LLMs), which has demonstrated superior generalization ability by balancing between larger overall model capacity and flexible per-instance specialization. Starting from a recent generalizable NeRF architecture called GNT, we first demonstrate that MoE can be neatly plugged in to enhance the model. We further customize a shared permanent expert and a geometry-aware consistency loss to enforce cross-scene consistency and spatial smoothness respectively, which are essential for generalizable view synthesis. Our proposed model, dubbed GNT with Mixture-of-View-Experts (GNT-MOVE), has experimentally shown state-of-the-art results when transferring to unseen scenes, indicating remarkably better cross-scene generalization in both zero-shot and few-shot settings. Our codes are available at https://github.com/VITA-Group/GNT-MOVE.

Zhiwen Fan, Jian Zhang, Renjie Li et al.

The rapid advancement of Large Multimodal Models (LMMs) for 2D images and videos has motivated extending these models to understand 3D scenes, aiming for human-like visual-spatial intelligence. Nevertheless, achieving deep spatial understanding comparable to human capabilities poses significant challenges in model encoding and data acquisition. Existing methods frequently depend on external depth sensors for geometry capture or utilize off-the-shelf algorithms for pre-constructing 3D maps, thereby limiting their scalability, especially with prevalent monocular video inputs and for time-sensitive applications. In this work, we introduce VLM-3R, a unified framework for Vision-Language Models (VLMs) that incorporates 3D Reconstructive instruction tuning. VLM-3R processes monocular video frames by employing a geometry encoder to derive implicit 3D tokens that represent spatial understanding. Leveraging our Spatial-Visual-View Fusion and over 200K curated 3D reconstructive instruction tuning question-answer (QA) pairs, VLM-3R effectively aligns real-world spatial context with language instructions. This enables monocular 3D spatial assistance and embodied reasoning. To facilitate the evaluation of temporal reasoning, we introduce the Vision-Spatial-Temporal Intelligence benchmark, featuring over 138.6K QA pairs across five distinct tasks focused on evolving spatial relationships. Extensive experiments demonstrate that our model, VLM-3R, not only facilitates robust visual-spatial reasoning but also enables the understanding of temporal 3D context changes, excelling in both accuracy and scalability.

Zhiwen Fan, Yifan Jiang, Peihao Wang et al.

Representing visual signals by implicit representation (e.g., a coordinate based deep network) has prevailed among many vision tasks. This work explores a new intriguing direction: training a stylized implicit representation, using a generalized approach that can apply to various 2D and 3D scenarios. We conduct a pilot study on a variety of implicit functions, including 2D coordinate-based representation, neural radiance field, and signed distance function. Our solution is a Unified Implicit Neural Stylization framework, dubbed INS. In contrary to vanilla implicit representation, INS decouples the ordinary implicit function into a style implicit module and a content implicit module, in order to separately encode the representations from the style image and input scenes. An amalgamation module is then applied to aggregate these information and synthesize the stylized output. To regularize the geometry in 3D scenes, we propose a novel self-distillation geometry consistency loss which preserves the geometry fidelity of the stylized scenes. Comprehensive experiments are conducted on multiple task settings, including novel view synthesis of complex scenes, stylization for implicit surfaces, and fitting images using MLPs. We further demonstrate that the learned representation is continuous not only spatially but also style-wise, leading to effortlessly interpolating between different styles and generating images with new mixed styles. Please refer to the video on our project page for more view synthesis results: https://zhiwenfan.github.io/INS.

Dejia Xu, Peihao Wang, Yifan Jiang et al.

Implicit Neural Representations (INRs) encoding continuous multi-media data via multi-layer perceptrons has shown undebatable promise in various computer vision tasks. Despite many successful applications, editing and processing an INR remains intractable as signals are represented by latent parameters of a neural network. Existing works manipulate such continuous representations via processing on their discretized instance, which breaks down the compactness and continuous nature of INR. In this work, we present a pilot study on the question: how to directly modify an INR without explicit decoding? We answer this question by proposing an implicit neural signal processing network, dubbed INSP-Net, via differential operators on INR. Our key insight is that spatial gradients of neural networks can be computed analytically and are invariant to translation, while mathematically we show that any continuous convolution filter can be uniformly approximated by a linear combination of high-order differential operators. With these two knobs, INSP-Net instantiates the signal processing operator as a weighted composition of computational graphs corresponding to the high-order derivatives of INRs, where the weighting parameters can be data-driven learned. Based on our proposed INSP-Net, we further build the first Convolutional Neural Network (CNN) that implicitly runs on INRs, named INSP-ConvNet. Our experiments validate the expressiveness of INSP-Net and INSP-ConvNet in fitting low-level image and geometry processing kernels (e.g. blurring, deblurring, denoising, inpainting, and smoothening) as well as for high-level tasks on implicit fields such as image classification.

Zhiwen Fan, Panwang Pan, Peihao Wang et al.

In the field of novel-view synthesis, the necessity of knowing camera poses (e.g., via Structure from Motion) before rendering has been a common practice. However, the consistent acquisition of accurate camera poses remains elusive, and errors in pose extraction can adversely impact the view synthesis process. To address this challenge, we introduce PF-GRT, a new Pose-Free framework for Generalizable Rendering Transformer, eliminating the need for pre-computed camera poses and instead leveraging feature-matching learned directly from data. PF-GRT is parameterized using a local relative coordinate system, where one of the source images is set as the origin. An OmniView Transformer is designed for fusing multi-view cues under the pose-free setting, where unposed-view fusion and origin-centric aggregation are performed. The 3D point feature along target ray is sampled by projecting onto the selected origin plane. The final pixel intensities are modulated and decoded using another Transformer. PF-GRT demonstrates an impressive ability to generalize to new scenes that were not encountered during the training phase, without the need of pre-computing camera poses. Our experiments with zero-shot rendering on the LLFF, RealEstate-10k, Shiny, and Blender datasets reveal that it produces superior quality in generating photo-realistic images. Moreover, it demonstrates robustness against noise in test camera poses. Code is available at https://zhiwenfan.github.io/PF-GRT/.

Dejia Xu, Yifan Jiang, Peihao Wang et al.

Virtual reality and augmented reality (XR) bring increasing demand for 3D content. However, creating high-quality 3D content requires tedious work that a human expert must do. In this work, we study the challenging task of lifting a single image to a 3D object and, for the first time, demonstrate the ability to generate a plausible 3D object with 360° views that correspond well with the given reference image. By conditioning on the reference image, our model can fulfill the everlasting curiosity for synthesizing novel views of objects from images. Our technique sheds light on a promising direction of easing the workflows for 3D artists and XR designers. We propose a novel framework, dubbed NeuralLift-360, that utilizes a depth-aware neural radiance representation (NeRF) and learns to craft the scene guided by denoising diffusion models. By introducing a ranking loss, our NeuralLift-360 can be guided with rough depth estimation in the wild. We also adopt a CLIP-guided sampling strategy for the diffusion prior to provide coherent guidance. Extensive experiments demonstrate that our NeuralLift-360 significantly outperforms existing state-of-the-art baselines. Project page: https://vita-group.github.io/NeuralLift-360/

Yi Wang, Zhiwen Fan, Tianlong Chen et al.

Vision Transformers (ViTs) have proven to be effective, in solving 2D image understanding tasks by training over large-scale image datasets; and meanwhile as a somehow separate track, in modeling the 3D visual world too such as voxels or point clouds. However, with the growing hope that transformers can become the "universal" modeling tool for heterogeneous data, ViTs for 2D and 3D tasks have so far adopted vastly different architecture designs that are hardly transferable. That invites an (over-)ambitious question: can we close the gap between the 2D and 3D ViT architectures? As a piloting study, this paper demonstrates the appealing promise to understand the 3D visual world, using a standard 2D ViT architecture, with only minimal customization at the input and output levels without redesigning the pipeline. To build a 3D ViT from its 2D sibling, we "inflate" the patch embedding and token sequence, accompanied with new positional encoding mechanisms designed to match the 3D data geometry. The resultant "minimalist" 3D ViT, named Simple3D-Former, performs surprisingly robustly on popular 3D tasks such as object classification, point cloud segmentation and indoor scene detection, compared to highly customized 3D-specific designs. It can hence act as a strong baseline for new 3D ViTs. Moreover, we note that pursing a unified 2D-3D ViT design has practical relevance besides just scientific curiosity. Specifically, we demonstrate that Simple3D-Former naturally enables to exploit the wealth of pre-trained weights from large-scale realistic 2D images (e.g., ImageNet), which can be plugged in to enhancing the 3D task performance "for free".

Zhiwen Fan, Peihao Wang, Yifan Jiang et al.

Neural volumetric representations have shown the potential that Multi-layer Perceptrons (MLPs) can be optimized with multi-view calibrated images to represent scene geometry and appearance, without explicit 3D supervision. Object segmentation can enrich many downstream applications based on the learned radiance field. However, introducing hand-crafted segmentation to define regions of interest in a complex real-world scene is non-trivial and expensive as it acquires per view annotation. This paper carries out the exploration of self-supervised learning for object segmentation using NeRF for complex real-world scenes. Our framework, called NeRF with Self-supervised Object Segmentation NeRF-SOS, couples object segmentation and neural radiance field to segment objects in any view within a scene. By proposing a novel collaborative contrastive loss in both appearance and geometry levels, NeRF-SOS encourages NeRF models to distill compact geometry-aware segmentation clusters from their density fields and the self-supervised pre-trained 2D visual features. The self-supervised object segmentation framework can be applied to various NeRF models that both lead to photo-realistic rendering results and convincing segmentation maps for both indoor and outdoor scenarios. Extensive results on the LLFF, Tank & Temple, and BlendedMVS datasets validate the effectiveness of NeRF-SOS. It consistently surpasses other 2D-based self-supervised baselines and predicts finer semantics masks than existing supervised counterparts. Please refer to the video on our project page for more details:https://zhiwenfan.github.io/NeRF-SOS.

Chenxin Li, Brandon Y. Feng, Zhiwen Fan et al.

Recent advances in neural rendering imply a future of widespread visual data distributions through sharing NeRF model weights. However, while common visual data (images and videos) have standard approaches to embed ownership or copyright information explicitly or subtly, the problem remains unexplored for the emerging NeRF format. We present StegaNeRF, a method for steganographic information embedding in NeRF renderings. We design an optimization framework allowing accurate hidden information extractions from images rendered by NeRF, while preserving its original visual quality. We perform experimental evaluations of our method under several potential deployment scenarios, and we further discuss the insights discovered through our analysis. StegaNeRF signifies an initial exploration into the novel problem of instilling customizable, imperceptible, and recoverable information to NeRF renderings, with minimal impact to rendered images. Project page: https://xggnet.github.io/StegaNeRF/.

Yunhao Yang, Junyuan Hong, Gabriel Jacob Perin et al.

Large language models (LLMs) can translate natural language instructions into executable action plans for robotics, autonomous driving, and other domains. Yet, deploying LLM-driven planning in the physical world demands strict adherence to safety and regulatory constraints, which current models often violate due to hallucination or weak alignment. Traditional data-driven alignment methods, such as Direct Preference Optimization (DPO), require costly human labeling, while recent formal-feedback approaches still depend on resource-intensive fine-tuning. In this paper, we propose LAD-VF, a fine-tuning-free framework that leverages formal verification feedback for automated prompt engineering. By introducing a formal-verification-informed text loss integrated with LLM-AutoDiff, LAD-VF iteratively refines prompts rather than model parameters. This yields three key benefits: (i) scalable adaptation without fine-tuning; (ii) compatibility with modular LLM architectures; and (iii) interpretable refinement via auditable prompts. Experiments in robot navigation and manipulation tasks demonstrate that LAD-VF substantially enhances specification compliance, improving success rates from 60% to over 90%. Our method thus presents a scalable and interpretable pathway toward trustworthy, formally-verified LLM-driven control systems.

Wenqing Zheng, S P Sharan, Zhiwen Fan et al.

Learning efficient and interpretable policies has been a challenging task in reinforcement learning (RL), particularly in the visual RL setting with complex scenes. While neural networks have achieved competitive performance, the resulting policies are often over-parameterized black boxes that are difficult to interpret and deploy efficiently. More recent symbolic RL frameworks have shown that high-level domain-specific programming logic can be designed to handle both policy learning and symbolic planning. However, these approaches rely on coded primitives with little feature learning, and when applied to high-dimensional visual scenes, they can suffer from scalability issues and perform poorly when images have complex object interactions. To address these challenges, we propose \textit{Differentiable Symbolic Expression Search} (DiffSES), a novel symbolic learning approach that discovers discrete symbolic policies using partially differentiable optimization. By using object-level abstractions instead of raw pixel-level inputs, DiffSES is able to leverage the simplicity and scalability advantages of symbolic expressions, while also incorporating the strengths of neural networks for feature learning and optimization. Our experiments demonstrate that DiffSES is able to generate symbolic policies that are simpler and more and scalable than state-of-the-art symbolic RL methods, with a reduced amount of symbolic prior knowledge.

Yuping Wang, Shuo Xing, Cui Can et al.

Generative Artificial Intelligence (GenAI) constitutes a transformative technological wave that reconfigures industries through its unparalleled capabilities for content creation, reasoning, planning, and multimodal understanding. This revolutionary force offers the most promising path yet toward solving one of engineering's grandest challenges: achieving reliable, fully autonomous driving, particularly the pursuit of Level 5 autonomy. This survey delivers a comprehensive and critical synthesis of the emerging role of GenAI across the autonomous driving stack. We begin by distilling the principles and trade-offs of modern generative modeling, encompassing VAEs, GANs, Diffusion Models, and Large Language Models (LLMs). We then map their frontier applications in image, LiDAR, trajectory, occupancy, video generation as well as LLM-guided reasoning and decision making. We categorize practical applications, such as synthetic data workflows, end-to-end driving strategies, high-fidelity digital twin systems, smart transportation networks, and cross-domain transfer to embodied AI. We identify key obstacles and possibilities such as comprehensive generalization across rare cases, evaluation and safety checks, budget-limited implementation, regulatory compliance, ethical concerns, and environmental effects, while proposing research plans across theoretical assurances, trust metrics, transport integration, and socio-technical influence. By unifying these threads, the survey provides a forward-looking reference for researchers, engineers, and policymakers navigating the convergence of generative AI and advanced autonomous mobility. An actively maintained repository of cited works is available at https://github.com/taco-group/GenAI4AD.

Xiangbo Gao, Runsheng Xu, Jiachen Li et al.

Perception is crucial for autonomous driving, but single-agent perception is often constrained by sensors' physical limitations, leading to degraded performance under severe occlusion, adverse weather conditions, and when detecting distant objects. Multi-agent collaborative perception offers a solution, yet challenges arise when integrating heterogeneous agents with varying model architectures. To address these challenges, we propose STAMP, a scalable task- and model-agnostic, collaborative perception pipeline for heterogeneous agents. STAMP utilizes lightweight adapter-reverter pairs to transform Bird's Eye View (BEV) features between agent-specific and shared protocol domains, enabling efficient feature sharing and fusion. This approach minimizes computational overhead, enhances scalability, and preserves model security. Experiments on simulated and real-world datasets demonstrate STAMP's comparable or superior accuracy to state-of-the-art models with significantly reduced computational costs. As a first-of-its-kind task- and model-agnostic framework, STAMP aims to advance research in scalable and secure mobility systems towards Level 5 autonomy. Our project page is at https://xiangbogaobarry.github.io/STAMP and the code is available at https://github.com/taco-group/STAMP.

Kairun Wen, Yuzhi Huang, Runyu Chen et al.

Understanding the dynamic physical world, characterized by its evolving 3D structure, real-world motion, and semantic content with textual descriptions, is crucial for human-agent interaction and enables embodied agents to perceive and act within real environments with human-like capabilities. However, existing datasets are often derived from limited simulators or utilize traditional Structurefrom-Motion for up-to-scale annotation and offer limited descriptive captioning, which restricts the capacity of foundation models to accurately interpret real-world dynamics from monocular videos, commonly sourced from the internet. To bridge these gaps, we introduce DynamicVerse, a physical-scale, multimodal 4D world modeling framework for dynamic real-world video. We employ large vision, geometric, and multimodal models to interpret metric-scale static geometry, real-world dynamic motion, instance-level masks, and holistic descriptive captions. By integrating window-based Bundle Adjustment with global optimization, our method converts long real-world video sequences into a comprehensive 4D multimodal format. DynamicVerse delivers a large-scale dataset consisting of 100K+ videos with 800K+ annotated masks and 10M+ frames from internet videos. Experimental evaluations on three benchmark tasks, namely video depth estimation, camera pose estimation, and camera intrinsics estimation, demonstrate that our 4D modeling achieves superior performance in capturing physical-scale measurements with greater global accuracy than existing methods.

Shijie Zhou, Haoran Chang, Sicheng Jiang et al.

3D scene representations have gained immense popularity in recent years. Methods that use Neural Radiance fields are versatile for traditional tasks such as novel view synthesis. In recent times, some work has emerged that aims to extend the functionality of NeRF beyond view synthesis, for semantically aware tasks such as editing and segmentation using 3D feature field distillation from 2D foundation models. However, these methods have two major limitations: (a) they are limited by the rendering speed of NeRF pipelines, and (b) implicitly represented feature fields suffer from continuity artifacts reducing feature quality. Recently, 3D Gaussian Splatting has shown state-of-the-art performance on real-time radiance field rendering. In this work, we go one step further: in addition to radiance field rendering, we enable 3D Gaussian splatting on arbitrary-dimension semantic features via 2D foundation model distillation. This translation is not straightforward: naively incorporating feature fields in the 3DGS framework encounters significant challenges, notably the disparities in spatial resolution and channel consistency between RGB images and feature maps. We propose architectural and training changes to efficiently avert this problem. Our proposed method is general, and our experiments showcase novel view semantic segmentation, language-guided editing and segment anything through learning feature fields from state-of-the-art 2D foundation models such as SAM and CLIP-LSeg. Across experiments, our distillation method is able to provide comparable or better results, while being significantly faster to both train and render. Additionally, to the best of our knowledge, we are the first method to enable point and bounding-box prompting for radiance field manipulation, by leveraging the SAM model. Project website at: https://feature-3dgs.github.io/

Rishov Sarkar, Hanxue Liang, Zhiwen Fan et al.

Computer vision researchers are embracing two promising paradigms: Vision Transformers (ViTs) and Multi-task Learning (MTL), which both show great performance but are computation-intensive, given the quadratic complexity of self-attention in ViT and the need to activate an entire large MTL model for one task. M$^3$ViT is the latest multi-task ViT model that introduces mixture-of-experts (MoE), where only a small portion of subnetworks ("experts") are sparsely and dynamically activated based on the current task. M$^3$ViT achieves better accuracy and over 80% computation reduction but leaves challenges for efficient deployment on FPGA. Our work, dubbed Edge-MoE, solves the challenges to introduce the first end-to-end FPGA accelerator for multi-task ViT with a collection of architectural innovations, including (1) a novel reordering mechanism for self-attention, which requires only constant bandwidth regardless of the target parallelism; (2) a fast single-pass softmax approximation; (3) an accurate and low-cost GELU approximation; (4) a unified and flexible computing unit that is shared by almost all computational layers to maximally reduce resource usage; and (5) uniquely for M$^3$ViT, a novel patch reordering method to eliminate memory access overhead. Edge-MoE achieves 2.24x and 4.90x better energy efficiency comparing with GPU and CPU, respectively. A real-time video demonstration is available online, along with our open-source code written using High-Level Synthesis.

Zhiwen Fan, Jian Zhang, Wenyan Cong et al.

Reconstructing and understanding 3D structures from a limited number of images is a well-established problem in computer vision. Traditional methods usually break this task into multiple subtasks, each requiring complex transformations between different data representations. For instance, dense reconstruction through Structure-from-Motion (SfM) involves converting images into key points, optimizing camera parameters, and estimating structures. Afterward, accurate sparse reconstructions are required for further dense modeling, which is subsequently fed into task-specific neural networks. This multi-step process results in considerable processing time and increased engineering complexity. In this work, we present the Large Spatial Model (LSM), which processes unposed RGB images directly into semantic radiance fields. LSM simultaneously estimates geometry, appearance, and semantics in a single feed-forward operation, and it can generate versatile label maps by interacting with language at novel viewpoints. Leveraging a Transformer-based architecture, LSM integrates global geometry through pixel-aligned point maps. To enhance spatial attribute regression, we incorporate local context aggregation with multi-scale fusion, improving the accuracy of fine local details. To tackle the scarcity of labeled 3D semantic data and enable natural language-driven scene manipulation, we incorporate a pre-trained 2D language-based segmentation model into a 3D-consistent semantic feature field. An efficient decoder then parameterizes a set of semantic anisotropic Gaussians, facilitating supervised end-to-end learning. Extensive experiments across various tasks show that LSM unifies multiple 3D vision tasks directly from unposed images, achieving real-time semantic 3D reconstruction for the first time.

Mukund Varma T, Peihao Wang, Zhiwen Fan et al.

In recent years, there has been an explosion of 2D vision models for numerous tasks such as semantic segmentation, style transfer or scene editing, enabled by large-scale 2D image datasets. At the same time, there has been renewed interest in 3D scene representations such as neural radiance fields from multi-view images. However, the availability of 3D or multiview data is still substantially limited compared to 2D image datasets, making extending 2D vision models to 3D data highly desirable but also very challenging. Indeed, extending a single 2D vision operator like scene editing to 3D typically requires a highly creative method specialized to that task and often requires per-scene optimization. In this paper, we ask the question of whether any 2D vision model can be lifted to make 3D consistent predictions. We answer this question in the affirmative; our new Lift3D method trains to predict unseen views on feature spaces generated by a few visual models (i.e. DINO and CLIP), but then generalizes to novel vision operators and tasks, such as style transfer, super-resolution, open vocabulary segmentation and image colorization; for some of these tasks, there is no comparable previous 3D method. In many cases, we even outperform state-of-the-art methods specialized for the task in question. Moreover, Lift3D is a zero-shot method, in the sense that it requires no task-specific training, nor scene-specific optimization.

Wenyan Cong, Hanqing Zhu, Kevin Wang et al.

Efficiently reconstructing 3D scenes from monocular video remains a core challenge in computer vision, vital for applications in virtual reality, robotics, and scene understanding. Recently, frame-by-frame progressive reconstruction without camera poses is commonly adopted, incurring high computational overhead and compounding errors when scaling to longer videos. To overcome these issues, we introduce VideoLifter, a novel video-to-3D pipeline that leverages a local-to-global strategy on a fragment basis, achieving both extreme efficiency and SOTA quality. Locally, VideoLifter leverages learnable 3D priors to register fragments, extracting essential information for subsequent 3D Gaussian initialization with enforced inter-fragment consistency and optimized efficiency. Globally, it employs a tree-based hierarchical merging method with key frame guidance for inter-fragment alignment, pairwise merging with Gaussian point pruning, and subsequent joint optimization to ensure global consistency while efficiently mitigating cumulative errors. This approach significantly accelerates the reconstruction process, reducing training time by over 82% while holding better visual quality than current SOTA methods.

Wenyan Cong, Hanqing Zhu, Peihao Wang et al.

World foundation models, which simulate the physical world by predicting future states from current observations and inputs, have become central to many applications in physical intelligence, including autonomous driving and robotics. However, these models require substantial computational resources for pretraining and are further constrained by available data during post-training. As such, scaling computation at test time emerges as both a critical and practical alternative to traditional model enlargement or re-training. In this work, we introduce SWIFT, a test-time scaling framework tailored for WFMs. SWIFT integrates our extensible WFM evaluation toolkit with process-level inference strategies, including fast tokenization, probability-based Top-K pruning, and efficient beam search. Empirical results on the COSMOS model demonstrate that test-time scaling exists even in a compute-optimal way. Our findings reveal that test-time scaling laws hold for WFMs and that SWIFT provides a scalable and effective pathway for improving WFM inference without retraining or increasing model size. Project page: https://scalingwfm.github.io/.

Longfei Li, Zhiwen Fan, Wenyan Cong et al.

Synthesizing realistic Martian landscape videos is crucial for mission rehearsal and robotic simulation. However, this task poses unique challenges due to the scarcity of high-quality Martian data and the significant domain gap between Martian and terrestrial imagery. To address these challenges, we propose a holistic solution composed of two key components: 1) A data curation pipeline Multimodal Mars Synthesis (M3arsSynth), which reconstructs 3D Martian environments from real stereo navigation images, sourced from NASA's Planetary Data System (PDS), and renders high-fidelity multiview 3D video sequences. 2) A Martian terrain video generator, MarsGen, which synthesizes novel videos visually realistic and geometrically consistent with the 3D structure encoded in the data. Our M3arsSynth engine spans a wide range of Martian terrains and acquisition dates, enabling the generation of physically accurate 3D surface models at metric-scale resolution. MarsGen, fine-tuned on M3arsSynth data, synthesizes videos conditioned on an initial image frame and, optionally, camera trajectories or textual prompts, allowing for video generation in novel environments. Experimental results show that our approach outperforms video synthesis models trained on terrestrial datasets, achieving superior visual fidelity and 3D structural consistency.

Wenyan Cong, Yiqing Liang, Yancheng Zhang et al.

Spatial intelligence, encompassing 3D reconstruction, perception, and reasoning, is fundamental to applications such as robotics, aerial imaging, and extended reality. A key enabler is the real-time, accurate estimation of core 3D attributes (camera parameters, point clouds, depth maps, and 3D point tracks) from unstructured or streaming imagery. Inspired by the success of large foundation models in language and 2D vision, a new class of end-to-end 3D geometric foundation models (GFMs) has emerged, directly predicting dense 3D representations in a single feed-forward pass, eliminating the need for slow or unavailable precomputed camera parameters. Since late 2023, the field has exploded with diverse variants, but systematic evaluation is lacking. In this work, we present the first comprehensive benchmark for 3D GFMs, covering five core tasks: sparse-view depth estimation, video depth estimation, 3D reconstruction, multi-view pose estimation, novel view synthesis, and spanning both standard and challenging out-of-distribution datasets. Our standardized toolkit automates dataset handling, evaluation protocols, and metric computation to ensure fair, reproducible comparisons. We evaluate 16 state-of-the-art GFMs, revealing their strengths and limitations across tasks and domains, and derive key insights to guide future model scaling and optimization. All code, evaluation scripts, and processed data will be publicly released to accelerate research in 3D spatial intelligence.

Zhiwen Fan, Panwang Pan, Peihao Wang et al.

Despite the significant progress in six degrees-of-freedom (6DoF) object pose estimation, existing methods have limited applicability in real-world scenarios involving embodied agents and downstream 3D vision tasks. These limitations mainly come from the necessity of 3D models, closed-category detection, and a large number of densely annotated support views. To mitigate this issue, we propose a general paradigm for object pose estimation, called Promptable Object Pose Estimation (POPE). The proposed approach POPE enables zero-shot 6DoF object pose estimation for any target object in any scene, while only a single reference is adopted as the support view. To achieve this, POPE leverages the power of the pre-trained large-scale 2D foundation model, employs a framework with hierarchical feature representation and 3D geometry principles. Moreover, it estimates the relative camera pose between object prompts and the target object in new views, enabling both two-view and multi-view 6DoF pose estimation tasks. Comprehensive experimental results demonstrate that POPE exhibits unrivaled robust performance in zero-shot settings, by achieving a significant reduction in the averaged Median Pose Error by 52.38% and 50.47% on the LINEMOD and OnePose datasets, respectively. We also conduct more challenging testings in causally captured images (see Figure 1), which further demonstrates the robustness of POPE. Project page can be found with https://paulpanwang.github.io/POPE/.