Recent advances in feed-forward 3D Gaussian Splatting have led to rapid improvements in efficient scene reconstruction from sparse views. However, most existing approaches construct Gaussian primitives directly aligned with the pixels in one or more of the input images. This leads to redundancies in the representation when input views overlap and constrains the position of the primitives to lie along the input rays without full flexibility in 3D space. Moreover, these pixel-aligned approaches do not naturally generalize to dynamic scenes, where effectively leveraging temporal information requires resolving both redundant and newly appearing content across frames. To address these limitations, we introduce a novel Fuse-and-Refine module that enhances existing feed-forward models by merging and refining the primitives in a canonical 3D space. At the core of our method is an efficient hybrid Splat-Voxel representation – from an initial set of pixel-aligned Gaussian primitives, we aggregate local features into a coarse-to-fine voxel hierarchy, and then use a sparse voxel transformer to process these voxel features and generate refined Gaussian primitives. By fusing and refining an arbitrary number of inputs into a consistent set of primitives, our representation effectively reduces redundancy and naturally adapts to temporal frames, enabling history-aware online reconstruction of dynamic scenes. Trained on large-scale static scene datasets, our model learns an effective global strategy to process around 200k primitives within 15ms and significantly enhances reconstruction quality compared to pixel-aligned reconstruction approaches. Without additional training, our model generalizes to video by fusing primitives across time, yielding a more temporally coherent result compared to baseline methods with graceful handling of occluded content. Our approach achieves state-of-the-art performance in both static and streaming scene reconstructions while running at interactive rates (15 fps with 350ms delay) on a single H100 GPU.

Large transformer models are proving to be a powerful tool for 3D vision and novel view synthesis. However, the standard Transformer’s well-known quadratic complexity makes it difficult to scale these methods to large scenes. To address this challenge, we propose the Local View Transformer (LVT), a large-scale scene reconstruction and novel view synthesis architecture that circumvents the need for the quadratic attention operation. Motivated by the insight that spatially nearby views provide more useful signal about the local scene composition than distant views, our model processes all information in a local neighborhood around each view. To attend to tokens in nearby views, we leverage a novel positional encoding that conditions on the relative geometric transformation between the query and nearby views. We decode the output of our model into a 3D Gaussian Splat scene representation that includes both color and opacity view-dependence. Taken together, the Local View Transformer enables reconstruction of arbitrarily large, high-resolution scenes in a single forward pass. See our project page for results and interactive demos: https://toobaimt.github.io/lvt/.

Humans judge perceptual similarity according to diverse visual attributes, including scene layout, subject location, and camera pose. Existing vision models understand a wide range of semantic abstractions but improperly weigh these attributes and thus make inferences misaligned with human perception. While vision representations have previously benefited from human preference alignment in contexts like image generation, the utility of perceptually aligned representations in more general-purpose settings remains unclear. Here, we investigate how aligning vision model representations to human perceptual judgments impacts their usability in standard computer vision tasks. We finetune state-of-the-art models on a dataset of human similarity judgments for synthetic image triplets and evaluate them across diverse computer vision tasks. We find that aligning models to perceptual judgments yields representations that improve upon the original backbones across many downstream tasks, including counting, semantic segmentation, depth estimation, instance retrieval, and retrieval-augmented generation. In addition, we find that performance is widely preserved on other tasks, including specialized out-of-distribution domains such as in medical imaging and 3D environment frames. Our results suggest that injecting an inductive bias about human perceptual knowledge into vision models can make them better representation learners.

We present a novel neural algorithm for performing high-quality, highresolution, real-time novel view synthesis. From a sparse set of input RGB images or videos streams, our network both reconstructs the 3D scene and renders novel views at 1080p resolution at 30fps on an NVIDIA A100. Our feed-forward network generalizes across a wide variety of datasets and scenes and produces state-of-the-art quality for a real-time method. Our quality approaches, and in some cases surpasses, the quality of some of the top offline methods. In order to achieve these results we use a novel combination of several key concepts, and tie them together into a cohesive and effective algorithm. We build on previous works that represent the scene using semi-transparent layers and use an iterative learned render-and-refine approach to improve those layers. Instead of flat layers, our method reconstructs layered depth maps (LDMs) that efficiently represent scenes with complex depth and occlusions. The iterative update steps are embedded in a multi-scale, UNet-style architecture to perform as much compute as possible at reduced resolution. Within each update step, to better aggregate the information from multiple input views, we use a specialized Transformer-based network component. This allows the majority of the per-input image processing to be performed in the input image space, as opposed to layer space, further increasing efficiency. Finally, due to the real-time nature of our reconstruction and rendering, we dynamically create and discard the internal 3D geometry for each frame, generating the LDM for each view. Taken together, this produces a novel and effective algorithm for view synthesis. Through extensive evaluation, we demonstrate that we achieve state-of-the-art quality at real-time rates.

Despite increasingly realistic image quality, recent 3D image generative models often operate on 3D volumes of fixed extent with limited camera motions. We investigate the task of unconditionally synthesizing unbounded nature scenes, enabling arbitrarily large camera motion while maintaining a persistent 3D world model. Our scene representation consists of an extendable, planar scene layout grid, which can be rendered from arbitrary camera poses via a 3D decoder and volume rendering, and a panoramic sky-dome. Based on this representation, we learn a generative world model solely from single-view internet photos. Our method enables simulating long flights through 3D landscapes, while maintaining global scene consistency-for instance, returning to the starting point yields the same view of the scene. Our approach enables scene extrapolation beyond the fixed bounds of current 3D generative models, while also supporting a persistent, camera-independent world representation that stands in contrast to auto-regressive 3D prediction models. Our project page: https://chail.github.io/persistent-nature/.

Current perceptual similarity metrics operate at the level of pixels and patches. These metrics compare images in terms of their low-level colors and textures, but fail to capture mid-level similarities and differences in image layout, object pose, and semantic content. In this paper, we develop a perceptual metric that assesses images holistically. Our first step is to collect a new dataset of human similarity judgments over image pairs that are alike in diverse ways. Critical to this dataset is that judgments are nearly automatic and shared by all observers. To achieve this we use recent text-to-image models to create synthetic pairs that are perturbed along various dimensions. We observe that popular perceptual metrics fall short of explaining our new data, and we introduce a new metric, DreamSim, tuned to better align with human perception. We analyze how our metric is affected by different visual attributes, and find that it focuses heavily on foreground objects and semantic content while also being sensitive to color and layout. Notably, despite being trained on synthetic data, our metric generalizes to real images, giving strong results on retrieval and reconstruction tasks. Furthermore, our metric outperforms both prior learned metrics and recent large vision models on these tasks. Our project page: https://dreamsim-nights.github.io/

Recent generative models can synthesize "views" of artificial images that mimic real-world variations, such as changes in color or pose, simply by learning from unlabeled image collections. Here, we investigate whether such views can be applied to real images to benefit downstream analysis tasks such as image classification. Using a pre-trained generator, we first find the latent code corresponding to a given real input image. Applying perturbations to the code creates natural variations of the image, which can then be ensembled together at test-time. We use StyleGAN2 as the source of generative augmentations and investigate this setup on classification tasks involving facial attributes, cat faces, and cars. Critically, we find that several design decisions are required towards making this process work; the perturbation procedure, weighting between the augmentations and original image, and training the classifier on synthesized images can all impact the result. Currently, we find that while test-time ensembling with GAN-based augmentations can offer some small improvements, the remaining bottlenecks are the efficiency and accuracy of the GAN reconstructions, coupled with classifier sensitivities to artifacts in GAN-generated images.