We present SAM 3D, a generative model for visually grounded 3D object reconstruction, predicting geometry, texture, and layout from a single image.
Personal Contribution: Main contributor to post-training, preference optimization.

We introduce Cue3D, the first comprehensive, model-agnostic framework for quantifying the influence of individual image cues in single-image 3D generation.

After training on large video data, will global 3D understanding naturally emerge? We study this by quantifying the 3D understanding of existing Video Foundation Models pretrained on vast video data, by estimating multiple 3D properties from their features via shallow read-outs.

We propose Reflect3D, a zero-shot single-image 3D reflection symmetry detector; we improve single-image 3D generation through symmetry-aware optimization.

We present a weakly-supervised Video State-Changing Object Segmentation (VSCOS) benchmark, revealing challenges in current VOS models for state-changing objects and introducing three solutions for improved state-changing object segmentation.

We introduce YouTubePD, the first public multimodal benchmark for Parkinson's Disease (PD) analysis, crowdsourced from existing YouTube videos featuring over 200 subjects.

We introduce a two-stage model and training strategies for one-shot object detection by integrating the metric learning with an anchor-free Faster R-CNN-style detection pipeline.

A few-shot segmentation dataset containing 1000 varied and balanced object categories with pixelwise annotation of ground-truth segmentation.