Artisan is a multi-agent system that adds physics-simulation feedback to articulated 3D asset generation, together with a benchmark for evaluating how generated objects look, move, and function.
Gen4P repurposes a video generator for feed-forward 4D perception, with single-pass encoding and continuous-time embeddings.
A lightweight refiner learns to correct the base model's velocity at each sampling step.
A plannable 4D world model that predicts future RGB video together with dense 4D geometry (depth, cameras, point clouds and scene flow) from an observed scene and text-specified actions.
A single feed-forward transformer that decomposes a whole-scene 3D latent into object-level meshes in one shared coordinate frame.
A pipeline to build a large-scale synthetic dataset for RGB(-D) hand pose estimation.
An agentic system that enables scalable generation of articulated 3D assets.
A feed-forward approach that utilizes kinematic prompt to guide 3D object articulation.
Holo4D re-films a dynamic video from a new camera path, using persistent 3D tracks as geometric control.
A large-scale synthetic multiview 4D dataset with dense geometry, tracking, camera motion, and human pose annotations.
A feed-forward approach that directly infers underlying articulated structure from a single static mesh
A video generator is reformulated as a 4D VAE to jointly reconstruct dense geometry and motion from monocular videos.
Given a monocular video, Mesh4D reconstructs a complete 3D mesh and tracks its motion across frames.
NOVA3R recovers complete, non-overlapping 3D geometry, reconstructing visible and occluded regions with physical plausibility.
A method to reconstruct compact and high-quality 3D Gaussians using Octree-based point fusion.
A method to repurpose video diffusion models for monocular 3D reconstruction of dynamic scenes.
A feed-forward approach to increase the likelihood that the 3D generator outputs stable 3D objects directly.
A model for completed 3D reconstruction from partially visible inputs.
An interactive video generative model that can serve as a motion prior for part-level dynamics.
Employing explicit correspondence matching as a geometry prior enables NeRF to generalize across scenes.
Semantic Style Transfer that simultaneously guides both style and appearance transfer.
A fast, super efficient, trainable on a single GPU in one day for scene 3D reconstruction from a single image.
A unified framework to recover better 2D appearance and 2.5D geometry.
A Feed-Forward model that reconstruct 3D structure and apperance with uncalibrated images.
Probe large vision models to determine to what extent they 'understand' different physical properties in an image.
A feed-forward approach for 360-degree scene-level novel view synthesis using only sparse observations.
A feed-forward approach for efficiently predicting 3D Gaussians from sparse multi-view images in a single forward pass.
A physical interaction with objects in vision for part-level dragging.
A self-organized 3D segmentation/decomposition model via neural implicit surface representation.
A method to synthesize consistent novel views from a single image on open-set categories without the need of explicit 3D representations.
Setting up a Stable Diffusion based network to solve the amodal completion problem for any category and without occluder mask provided.
A versatile plug-and-play module to fix the scheduler flaws for diffusion models.
An indoor panorama outpainting model using latent diffusion models with view-consistent.
PICFormer achieves pluralistic image completion with multiple and diverse solutions using a transformer based architecture.
A generalized framework for multi-modality control based on text-to-image generation.
A simple approach to avoid codebook collapse and achive 100% codebook utilisation.
Minimize the codebook-data distortion as the Wasserstein distance.
A unified discrete diffusion model for simultaneous vision-language generation.
A spatially conditional normalization is introduced to address the repeated artifacts in vector quantized methods.
Automatically decompose a scene into 3D instance, trained using only 2D semantic lables and images.
A model that transfers the 2D semantic map into 3D NeRF, and lets users edit 3D model through 2D semantic input.
TFill fills in reasonable contents for both foreground object removal and content completion.
A high-level scene understanding system that simultaneously models the completed shape and appearance for all instances.
A GAN inversion model is trained for Stylizing Portraits.
A novel spatially-correlative loss that is simple, efficient and yet effective for preserving scene structure consistency while supporting large appearance changes during unpaired I2I translation.
Given a masked image, the proposed pic model is able to generate multiple and diverse plausible results.
Without using any real depth map, the proposed model evaluates depth maps on real scenes using only synthetic datasets.
arxiv