3D reconstruction and view synthesis are foundational problems in computer vision, graphics, and immersive technologies such as augmented reality (AR), virtual reality (VR), and digital twins. Traditional methods rely on computationally intensive iterative optimization in a complex chain, limiting their applicability in real-world scenarios. Recent advances in feed-forward approaches, driven by deep learning, have revolutionized this field by enabling fast and generalizable 3D reconstruction and view synthesis. This survey offers a comprehensive review of feed-forward techniques for 3D reconstruction and view synthesis, with a taxonomy according to the underlying representation architectures including point cloud, 3D Gaussian Splatting (3DGS), Neural Radiance Fields (NeRF), etc. We examine key tasks such as pose-free reconstruction, dynamic 3D reconstruction, and 3D-aware image and video synthesis, highlighting their applications in digital humans, SLAM, robotics, and beyond. In addition, we review commonly used datasets with detailed statistics, along with evaluation protocols for various downstream tasks. We conclude by discussing open research challenges and promising directions for future work, emphasizing the potential of feed-forward approaches to advance the state of the art in 3D vision. The project page is available at https://fnzhan.com/projects/Feed-Forward-3D Feed-Forward-3D.

Self-distillation bootstraps large language models (LLMs) by training on their own generations. However, existing methods either rely on external signals to curate self-generated outputs (e.g., correctness filtering, execution feedback, and reward search), which are costly and unavailable for the best-performing frontier models, or skip curation entirely and train on all raw outputs, an approach that is often domain-specific and hard to generalize. Both also share a deeper weakness that self-generated outputs entangle task-relevant capability with others, such as stylistic patterns, formatting artifacts, and model-specific errors, diluting the signal for the specific capability one aims to improve. In this paper, we propose Self-Policy Distillation (SPD), which achieves generalizable, capability selective without any external signal. Specifically, SPD extracts a low-rank capability subspace from the model's own gradients on correctness-defining tokens, projects key-value (KV) activations into this subspace during self-generation, and fine-tunes on the resulting raw outputs with standard next-token prediction loss. Through extensive experiments across code generation, mathematical reasoning, and multiple-choice QA, we show that SPD achieves up to 13

The advancements in diffusion models for 2D and 3D content creation have sparked a surge of interest in generating 4D content. However, the scarcity of 3D scene datasets constrains current methodologies to primarily object-centric generation. To overcome this limitation, we present Comp4D, a novel framework for text-to-compositional 4D scene generation. Unlike conventional methods that generate a singular 4D representation of the entire scene, Comp4D innovatively employs a decompose-then-recompose strategy, constructing each 4D component within the scene separately. The framework first decomposes a textual input prompt into multiple object components and delineates their moving trajectories. After initializing the static 3D objects, we construct the compositional 4D scene by accurately positioning these objects along their designated paths. To refine the scene and motion, our method proposes a novel compositional score distillation technique involving trajectory-guided and object-centric sampling, utilizing pre-trained diffusion models across text-to-image, text-to-video, and text-to-3D domains for optimization. Extensive experiments demonstrate our superior 4D content creation capability compared to prior arts, showcasing superior visual quality, motion fidelity, and enhanced object interactions.

Recent advancements in static feed-forward scene reconstruction have demonstrated significant progress in high-quality novel view synthesis. However, these models often struggle with generalizability across diverse environments and fail to effectively handle dynamic content. We present BTimer (short for BulletTimer), the first motion-aware feed-forward model for real-time reconstruction and novel view synthesis of dynamic scenes. Our approach reconstructs the full scene in a 3D Gaussian Splatting representation at a given target ('bullet') timestamp by aggregating information from all the context frames. Such a formulation allows BTimer to gain scalability and generalization by leveraging both static and dynamic scene datasets. Given a casual monocular dynamic video, BTimer reconstructs a bullet-time scene within 150ms while reaching state-of-the-art performance on both static and dynamic scene datasets, even compared with optimization-based approaches.

Implicit surface representations such as the signed distance function (SDF) have emerged as a promising approach for image-based surface reconstruction. However, existing optimization methods assume solid surfaces and are therefore unable to properly reconstruct semi-transparent surfaces and thin structures, which also exhibit low opacity due to the blending effect with the background. While neural radiance field (NeRF) based methods can model semi-transparency and achieve photo-realistic quality in synthesized novel views, their volumetric geometry representation tightly couples geometry and opacity, and therefore cannot be easily converted into surfaces without introducing artifacts. We present $\alpha$Surf, a novel surface representation with decoupled geometry and opacity for the reconstruction of semi-transparent and thin surfaces where the colors mix. Ray-surface intersections on our representation can be found in closed-form via analytical solutions of cubic polynomials, avoiding Monte-Carlo sampling and is fully differentiable by construction. Our qualitative and quantitative evaluations show that our approach can accurately reconstruct surfaces with semi-transparent and thin parts with fewer artifacts, achieving better reconstruction quality than state-of-the-art SDF and NeRF methods. Website: https://alphasurf.netlify.app/

Recent advancements in diffusion models for 2D and 3D content creation have sparked a surge of interest in generating 4D content. However, the scarcity of 3D scene datasets constrains current methodologies to primarily object-centric generation. To overcome this limitation, we present Comp4D, a novel framework for Compositional 4D Generation. Unlike conventional methods that generate a singular 4D representation of the entire scene, Comp4D innovatively constructs each 4D object within the scene separately. Utilizing Large Language Models (LLMs), the framework begins by decomposing an input text prompt into distinct entities and maps out their trajectories. It then constructs the compositional 4D scene by accurately positioning these objects along their designated paths. To refine the scene, our method employs a compositional score distillation technique guided by the pre-defined trajectories, utilizing pre-trained diffusion models across text-to-image, text-to-video, and text-to-3D domains. Extensive experiments demonstrate our outstanding 4D content creation capability compared to prior arts, showcasing superior visual quality, motion fidelity, and enhanced object interactions.

The availability of large-scale multimodal datasets and advancements in diffusion models have significantly accelerated progress in 4D content generation. Most prior approaches rely on multiple images or video diffusion models, utilizing score distillation sampling for optimization or generating pseudo novel views for direct supervision. However, these methods are hindered by slow optimization speeds and multi-view inconsistency issues. Spatial and temporal consistency in 4D geometry has been extensively explored respectively in 3D-aware diffusion models and traditional monocular video diffusion models. Building on this foundation, we propose a strategy to migrate the temporal consistency in video diffusion models to the spatial-temporal consistency required for 4D generation. Specifically, we present a novel framework, \textbf{Diffusion4D}, for efficient and scalable 4D content generation. Leveraging a meticulously curated dynamic 3D dataset, we develop a 4D-aware video diffusion model capable of synthesizing orbital views of dynamic 3D assets. To control the dynamic strength of these assets, we introduce a 3D-to-4D motion magnitude metric as guidance. Additionally, we propose a novel motion magnitude reconstruction loss and 3D-aware classifier-free guidance to refine the learning and generation of motion dynamics. After obtaining orbital views of the 4D asset, we perform explicit 4D construction with Gaussian splatting in a coarse-to-fine manner. Extensive experiments demonstrate that our method surpasses prior state-of-the-art techniques in terms of generation efficiency and 4D geometry consistency across various prompt modalities.

Cross-scene generalizable NeRF models, which could directly synthesize novel views using several source views of unseen scenes, are gaining prominence in the NeRF field. Discovering the potential signal of emerging capabilities in existing methods, we draw a parallel between BERT’s "drop-and-predict" Masked Language Model (MLM) pre-training and novel view synthesis (NVS) in generalizable NeRF. In this work, we pioneer the scaling up of NVS as an effective pretraining strategy in a multi-view context. To bolster generalizability in pretraining, we incorporate a large-scale, minimally annotated dataset and proportionally increase the model size, revealing a neural scaling law akin to that observed in BERT. We also introduce innovative hardness-aware training techniques to enhance robust feature learning. Our model, named "NPS", demonstrates remarkable generalizability in both zero-shot and few-shot novel view synthesis. It further shows emergent capabilities in downstream tasks like few-shot multi-view semantic segmentation and depth estimation. Significantly, NPS reduces the necessity of training separate models for each task, underlining its versatility and efficiency. This approach sets a new precedent in the NeRF field, and highlights the vast possibilities opened up by scaling up generalizable novel view synthesis.

We present SCube, a novel method for reconstructing large-scale 3D scenes (geometry, appearance, and semantics) from a sparse set of posed images. Our method encodes reconstructed scenes using a novel representation VoxSplat, which is a set of 3D Gaussians supported on a high-resolution sparse-voxel scaffold. To reconstruct a VoxSplat from images, we employ a hierarchical voxel latent diffusion model conditioned on the input images followed by a feedforward appearance prediction model. The diffusion model generates high-resolution grids progressively in a coarse-to-fine manner, and the appearance network predicts a set of Gaussians within each voxel. From as few as 3 non-overlapping input images, SCube can generate millions of Gaussians with a 1024^3 voxel grid spanning hundreds of meters in 20 seconds. Past works tackling scene reconstruction from images either rely on per-scene optimization and fail to reconstruct the scene away from input views (thus requiring dense view coverage as input) or leverage geometric priors based on low-resolution models, which produce blurry results. In contrast, SCube leverages high-resolution sparse networks and produces sharp outputs from few views. We show the superiority of SCube compared to prior art using the Waymo self-driving dataset on 3D reconstruction and demonstrate its applications, such as LiDAR simulation and text-to-scene generation.

The landscape of computer graphics has undergone significant transformations with the recent advances of differentiable rendering models. These rendering models often rely on heuristic designs that may not fully align with the final rendering objectives. We address this gap by pioneering \textit{evolutive rendering models}, a methodology where rendering models possess the ability to evolve and adapt dynamically throughout the rendering process. In particular, we present a comprehensive learning framework that enables the optimization of three principal rendering elements, including the gauge transformations, the ray sampling mechanisms, and the primitive organization. Central to this framework is the development of differentiable versions of these rendering elements, allowing for effective gradient backpropagation from the final rendering objectives. A detailed analysis of gradient characteristics is performed to facilitate a stable and goal-oriented elements evolution. Our extensive experiments demonstrate the large potential of evolutive rendering models for enhancing the rendering performance across various domains, including static and dynamic scene representations, generative modeling, and texture mapping.

We present L4GM, the first 4D Large Reconstruction Model that produces animated objects from a single-view video input -- in a single feed-forward pass that takes only a second. Key to our success is a novel dataset of multiview videos containing curated, rendered animated objects from Objaverse. This dataset depicts 44K diverse objects with 110K animations rendered in 48 viewpoints, resulting in 12M videos with a total of 300M frames. We keep our L4GM simple for scalability and build directly on top of LGM, a pretrained 3D Large Reconstruction Model that outputs 3D Gaussian ellipsoids from multiview image input. L4GM outputs a per-frame 3D Gaussian splat representation from video frames sampled at a low fps and then upsamples the representation to a higher fps to achieve temporal smoothness. We add temporal self-attention layers to the base LGM to help it learn consistency across time, and utilize a per-timestep multiview rendering loss to train the model. The representation is upsampled to a higher framerate by training an interpolation model which produces intermediate 3D Gaussian representations. We showcase that L4GM that is only trained on synthetic data generalizes well on in-the-wild videos, producing high quality animated 3D assets.

Neural view synthesis (NVS) is one of the most successful techniques for synthesizing free viewpoint videos, capable of achieving high fidelity from only a sparse set of captured images. This success has led to many variants of the techniques, each evaluated on a set of test views typically using image quality metrics such as PSNR, SSIM, or LPIPS. There has been a lack of research on how NVS methods perform with respect to perceived video quality. We present the first study on perceptual evaluation of NVS and NeRF variants. For this study, we collected two datasets of scenes captured in a controlled lab environment as well as in-the-wild. In contrast to existing datasets, these scenes come with reference video sequences, allowing us to test for temporal artifacts and subtle distortions that are easily overlooked when viewing only static images. We measured the quality of videos synthesized by several NVS methods in a well-controlled perceptual quality assessment experiment as well as with many existing state-of-the-art image/video quality metrics. We present a detailed analysis of the results and recommendations for dataset and metric selection for NVS evaluation.

The computer vision community is embracing two promising learning paradigms: the Vision Transformer (ViT) and Multi-task Learning (MTL). ViT models show extraordinary performance over traditional convolution networks but are commonly recognized as computation-intensive, especially the self-attention with quadratic complexity. MTL uses one model to infer multiple tasks with better performance by enforcing shared representation among tasks, but a huge drawback is that, most MTL regimes require activation of the entire model even when only one or a few tasks are needed, causing significant computing waste. M 3 ViT is the latest multi-task Vi $T$ 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 Vi $T$ achieves better accuracy and over 80% computation reduction and paves the way for efficient real-time MTL using ViT. Despite the algorithmic advantages of MTL, ViT, and even M 3 ViT, there are still many challenges for efficient deployment on FPGA. For instance, in general Transformer/ViT models, the self-attention is known as computational intensive and requires high bandwidth. In addition, softmax operations and the activation function GELU are extensively used, which unfortunately can consume more than half of the entire FPGA resource (LUTs). In the M 3 ViT model, the promising MoE mechanism for multi-task exposes new challenges for memory access overhead and also increases resource usage because of more layer types. To address these challenges in both general Transformer/ViT models and the state-of-the-art multi-task M 3 ViT with MoE, we propose Edge-MoE, the first end-to-end FPGA accelerator for multi-task ViT with a rich collection of architectural innovations. First, for general Transformer/ViT models, we propose (1) a novel reordering mechanism for self-attention, which reduces the bandwidth requirement from proportional to constant regardless of the target parallelism; (2) a fast single-pass softmax approximation; (3) an accurate and low-cost GELU approximation, which can significantly reduce the computation latency and resource usage; and (4) a unified and flexible computing unit that can be shared by almost all computational layers to maximally reduce resource usage. Second, for the advanced multi-task M 3 ViT with MoE, we propose a novel patch reordering method to completely eliminate any memory access overhead. Third, we deliver on-board implementation and measurement on Xilinx ZCU102 FPGA, with verified functionality and open-sourced hardware design, which achieves 2.24× and 4.90× better energy efficiency comparing with GPU (A6000) and CPU (Xeon 6226R), respectively. A real-time video demonstration of our accelerated multi-task ViT on an autonomous driving dataset is available in GitHub, 1 1 https://github.com/sharc-lab/Edge-MoE/raw/main/demo.mp4 together with our FPGA design using High-Level Synthesis, host code, FPGA bitstream, and on-board performance results.

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 [52], 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.

The superiority of deep learning based point cloud representations relies on large-scale labeled datasets, while the annotation of point clouds is notoriously expensive. One of the most effective solutions is to transfer the knowledge from existing labeled source data to unlabeled target data. However, domain bias typically hinders knowledge transfer and leads to accuracy degradation. In this paper, we propose a Masked Local Structure Prediction (MLSP) method to encode target data. Along with the supervised learning on the source domain, our method enables models to embed source and target data in a shared feature space. Specifically, we predict masked local structure via estimating point cardinality, position and normal. Our design philosophies lie in: 1) Point cardinality reflects basic structures (e.g., line, edge and plane) that are invariant to specific domains. 2) Predicting point positions in masked areas generalizes learned representations so that they are robust to incompletion-caused domain bias. 3) Point normal is generated by neighbors and thus robust to noise across domains. We conduct experiments on shape classification and semantic segmentation with different transfer permutations and the results demonstrate the effectiveness of our method. Code is available at https://github.com/VITA-Group/MLSP .

In both classification and segmentation experiments, all the network configurations are the same except for the supervision head. We use DGCNN [4] as the shared feature extractor, which is composed of four point cloud convolution layers of sizes [64, 64, 128, 256] and 1D convolution layer with kernel size 1 and channel number 1024. The output is max-pooled to get global feature. For SSL heads, both normal prediction hnorm and position prediction head hpos are implemented similarly with four 1D convolution layers of sizes [256, 256, 128, 3]. The cardinality prediction head is implemented with a 1D convolution layer of size 512 and three fully connected layers of sizes [256, 256, 8] (where 8 is the number of cardinality classes). In PointDA-10 benchmark, the classification head hsup is implemented using three fully connected layers with sizes [512, 256, 10] and takes as input the global feature vector. In PointSegDA benchmark, the segmentation head takes the global feature vector concatenated to feature representations of each point, and it is implemented using four 1D convolution layers of sizes [256, 256, 128, 8]. In all the heads, non-linearity is ReLU and a dropout of 0.5 is applied to the first two hidden layers. Batch normalization is applied after all convolution layers.

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.

Current 3D object detection paradigms highly rely on extensive annotation efforts, which makes them not practical in many real-world industrial applications. Inspired by that a human driver can keep accumulating experiences from self-exploring the roads without any tutor's guidance, we first step forwards to explore a simple yet effective self-supervised learning framework tailored for LiDAR-based 3D object detection. Although the self-supervised pipeline has achieved great success in 2D domain, the characteristic challenges (e.g., complex geometry structure and various 3D object views) encountered in the 3D domain hinder the direct adoption of existing techniques that often contrast the 2D augmented data or cluster single-view features. Here we present a novel self-supervised 3D Object detection framework that seamlessly integrates the geometry-aware contrast and clustering harmonization to lift the unsupervised 3D representation learning, named GCC-3D. First, GCC-3D introduces a Geometric-Aware Contrastive objective to learn spatial-sensitive local structure representation. This objective enforces the spatially close voxels to have high feature similarity. Second, a Pseudo-Instance Clustering harmonization mechanism is proposed to encourage that different views of pseudo-instances should have consistent similarities to clustering prototype centers. This module endows our model semantic discriminative capacity. Extensive experiments demonstrate our GCC-3D achieves significant performance improvement on data-efficient 3D object detection benchmarks (nuScenes and Waymo). Moreover, our GCC-3D framework can achieve state-of-the art performances on all popular 3D object detection benchmarks.

Current perception models in autonomous driving have become notorious for greatly relying on a mass of annotated data to cover unseen cases and address the long-tail problem. On the other hand, learning from unlabeled large-scale collected data and incrementally self-training powerful recognition models have received increasing attention and may become the solutions of next-generation industry-level powerful and robust perception models in autonomous driving. However, the research community generally suffered from data inadequacy of those essential real-world scene data, which hampers the future exploration of fully/semi/self-supervised methods for 3D perception. In this paper, we introduce the ONCE (One millioN sCenEs) dataset for 3D object detection in the autonomous driving scenario. The ONCE dataset consists of 1 million LiDAR scenes and 7 million corresponding camera images. The data is selected from 144 driving hours, which is 20x longer than the largest 3D autonomous driving dataset available (e.g. nuScenes and Waymo), and it is collected across a range of different areas, periods and weather conditions. To facilitate future research on exploiting unlabeled data for 3D detection, we additionally provide a benchmark in which we reproduce and evaluate a variety of self-supervised and semi-supervised methods on the ONCE dataset. We conduct extensive analyses on those methods and provide valuable observations on their performance related to the scale of used data. Data, code, and more information are available at https://once-for-auto-driving.github.io/index.html.