Jaesik Park

Direct Reward Fine-Tuning on Poses for Single Image to 3D Human in the Wild

TRACE: Your Diffusion Model is Secretly an Instance Edge Detector

MotionStream: Real-Time Video Generation with Interactive Motion Controls

Optimized Minimal 4D Gaussian Splatting

Occluded 3D Object Reconstruction via Masked Multi-view Volumetric Transformer

Subject-driven Video Generation Emerges from Experience Replays

Lightweight Image-to-3D Shape Generation via Vitality-Aware Pruning and Quantization

JAM-Flow: Joint Audio-Motion Synthesis with Flow Matching

Enhancing structural consistency of 3D Human Pose Estimation through Trainable Loss Function

Alleviating Suboptimality of Flow Maps with Improved Self-Distillation Guidance

Extend3D: Town-scale 3D Generation

OpenBox: Annotate Any Bounding Boxes in 3D

Tree-Guided Diffusion Planner

Holistic Order Prediction in Natural Scenes

Metropolis-Hastings Sampling for 3D Gaussian Reconstruction

Designing Concise ConvNets with Columnar Stages

Locality-aware Gaussian Compression for Fast and High-quality Rendering

Cross Resolution Encoding-Decoding For Detection Transformers

Improving Editability in Compositional Image Diffusion with Layer-wise Memory