We introduce Cosmos-Transfer1, a conditional world generation model that can generate world simulations based on multiple spatial control inputs of various modalities such as segmentation, depth, and edge. In the design, the spatial conditional scheme is adaptive and customizable. It allows weighting different conditional inputs differently at different spatial locations. This enables highly controllable world generation and finds use in various world-to-world transfer use cases, including Sim2Real. We conduct extensive evaluations to analyze the proposed model and demonstrate its applications for Physical AI, including robotics Sim2Real and autonomous vehicle data enrichment. We further demonstrate an inference scaling strategy to achieve real-time world generation with an NVIDIA GB200 NVL72 rack.

Previous methods for scene generation either suffer from limited scales or lack geometric and appearance Generating simulatable and controllable 3D scenes is an essential consistency along generated sequences. In contrast, task for a wide spectrum of applications, including we leverage the recent advancements in scalable 3D mixed reality, robotics, and the training and testing of autonomous representation and video models to achieve large dynamic vehicles (AV) [25, 33]. In particular, the requirements scene generation that allows flexible controls through HD of AV applications have introduced new challenges maps, vehicle bounding boxes, and text descriptions. First, for 3D generative models in driving scenarios, posing the we construct a map-conditioned sparse-voxel-based 3D following key desiderata: (1) fidelity and consistency, to generative model to unleash its power for unbounded voxel ensure that the generated scenes support photo-realistic rendering world generation. Then, we re-purpose a video model and while preserving consistent appearance and geometry ground it on the voxel world through a set of carefully designed for reliable and stable physics simulation; (2) largescale, pixel-aligned guidance buffers, synthesizing a consistent to generate scenes at map-level for traffic simulation; appearance. Finally, we propose a fast feed-forward and (3) controllability, to allow easy manipulation of the approach that employs both voxel and pixel branches to lift scene layout, appearance, and ego-car behaviors for curating the dynamic videos to dynamic 3D Gaussians with control-adversarial scenarios.

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.

We present fVDB, a novel GPU-optimized framework for deep learning on large-scale 3D data. fVDB provides a complete set of differentiable primitives to build deep learning architectures for common tasks in 3D learning such as convolution, pooling, attention, ray-tracing, meshing, etc. fVDB simultaneously provides a much larger feature set (primitives and operators) than established frameworks with no loss in efficiency: our operators match or exceed the performance of other frameworks with narrower scope. Furthermore, fVDB can process datasets with much larger footprint and spatial resolution than prior works, while providing a competitive memory footprint on small inputs. To achieve this combination of versatility and performance, fVDB relies on a single novel VDB index grid acceleration structure paired with several key innovations including GPU accelerated sparse grid construction, convolution using tensorcores, fast ray tracing kernels using a Hierarchical Digital Differential Analyzer algorithm (HDDA), and jagged tensors. Our framework is fully integrated with PyTorch enabling interoperability with existing pipelines, and we demonstrate its effectiveness on a number of representative tasks such as large-scale point-cloud segmentation, high resolution 3D generative modeling, unbounded scale Neural Radiance Fields, and large-scale point cloud reconstruction.

We present $\mathcal{X}^3$ (pronounced XCube), a novel generative model for high-resolution sparse 3D voxel grids with arbitrary attributes. Our model can generate millions of voxels with a finest effective resolution of up to $1024^3$ in a feed-forward fashion without time-consuming test-time optimization. To achieve this, we employ a hierarchical voxel latent diffusion model which generates progressively higher resolution grids in a coarse-to-fine manner using a custom framework built on the highly efficient VDB data structure. Apart from generating high-resolution objects, we demonstrate the effectiveness of XCube on large outdoor scenes at scales of 100m$\times$100m with a voxel size as small as 10cm. We observe clear qualitative and quantitative improvements over past approaches. In addition to unconditional generation, we show that our model can be used to solve a variety of tasks such as user-guided editing, scene completion from a single scan, and text-to-3D. More results and details can be found at https://research.nvidia.com/labs/toronto-ai/xcube/.

Disentangled generative models are typically trained with an extra regularization term, which encourages the traversal of each latent factor to make a distinct and independent change at the cost of generation quality. When traversing the latent space of generative models trained without the disentanglement term, the generated samples show semantically meaningful change, raising the question: do generative models know disentanglement? We propose an unsupervised and model-agnostic method: Disentanglement via Contrast (DisCo) in the Variation Space. DisCo consists of: (i) a Navigator providing traversal directions in the latent space, and (ii) a $\Delta$-Contrastor composed of two shared-weight Encoders, which encode image pairs along these directions to disentangled representations respectively, and a difference operator to map the encoded representations to the Variation Space. We propose two more key techniques for DisCo: entropy-based domination loss to make the encoded representations more disentangled and the strategy of flipping hard negatives to address directions with the same semantic meaning. By optimizing the Navigator to discover disentangled directions in the latent space and Encoders to extract disentangled representations from images with Contrastive Learning, DisCo achieves the state-of-the-art disentanglement given pretrained non-disentangled generative models, including GAN, VAE, and Flow. Project page at https://github.com/xrenaa/DisCo.

Content and style (C-S) disentanglement intends to decompose the underlying explanatory factors of objects into two independent subspaces. From the unsupervised disentanglement perspective, we rethink content and style and propose a formulation for unsupervised C-S disentanglement based on our assumption that different factors are of different importance and popularity for image reconstruction, which serves as a data bias. The corresponding model inductive bias is introduced by our proposed C-S disentanglement Module (C-S DisMo), which assigns different and independent roles to content and style when approximating the real data distributions. Specifically, each content embedding from the dataset, which encodes the most dominant factors for image reconstruction, is assumed to be sampled from a shared distribution across the dataset. The style embedding for a particular image, encoding the remaining factors, is used to customize the shared distribution through an affine transformation. The experiments on several popular datasets demonstrate that our method achieves the state-of-the-art unsupervised C-S disentanglement, which is comparable or even better than supervised methods. We verify the effectiveness of our method by downstream tasks: domain translation and single-view 3D reconstruction. Project page at https://github.com/xrenaa/CS-DisMo.