WonderLand is a video-latent based approach for single-image 3D reconstruction in large-scale scenes.

Omni-ID is a novel facial representation tailored for generative tasks, encoding identity features from unstructured images into a fixed-size representation that captures diverse expressions and poses.

AC3D studies when and how you should condition camera signals into a video diffusion model for a better camera control and a higher video quality.

AToM trains a single text-to-mesh model on many prompts using 2D diffusion without 3D supervision, yileds high-quality textured meshes under a second, and generalizes to unseen prompts.

Magic123 is a coarse-to-fine image-to-3D pipeline that produces high-quality high-resolution 3D content from a single unposed image by the guidance of both 2D and 3D priors.

We propose a state-of-the-art Standard Transformer model for point cloud understanding and find that image pretraining helps point cloud tasks.

ZeroSeg trains open-vocabulary zero-shot semantic segmentation models using only CLIP Vision Encoder

PointNeXt boosts the performance of PointNet++ to the state-of-the-art level with improved training and scaling strategies.

ASSANet makes PointNet++ faster and more accurate.

This work transfers concepts such as residual/dense connections and dilated convolutions from CNNs to GCNs in order to successfully train very deep GCNs.