Jiwen Lu

Full Professor@Tsinghua University·中国·OpenReview

研究方向

Neural Networks Architecture · Visual Editing and Forensics · 3D Vision · Human Action Recognition · Unsupervised Representation Learning · Deep Metric Learning · Human Face Analysis · Image and Video Search · Deep Reinforcement Learning · Visual Tracking and Re-Identification

Jiwen Lu

Streaming Visual Geometry Transformer

Latent Diffusion Model without Variational Autoencoder

Improving Autoregressive Video Modeling with History Understanding

R2RGEN: Real-to-Real 3D Data Generation for Spatially Generalized Manipulation

DiffMoE: Dynamic Token Selection for Scalable Diffusion Transformers

Underwater Visual Geometry Estimation with Self-supervised Prototype-Graph Modulation

Astra: General Interactive World Model with Autoregressive Denoising

VARestorer: One-Step VAR Distillation for Real-World Image Super-Resolution

EmbodiedGS: Reconstruct Unified Embodied Representation from RGB Stream

MoRe4D: Joint 3D Motion Generation and Geometry Reconstruction for 4D Synthesis from a Single Image

SparseCodeQ: Extreme Sparse Coding Quantization for Large Vision-Language Models

MVP: Memory-enhanced Vision-Language-Action Policy with Feedback Learning

Lightning Video: Building Compact Diffusion Transformers for High-Fidelity On-Device Video Generation

Quantize-then-Rectify: Accelerating VQ-VAE Training in Latent Feature Space

Terra: Explorable Native 3D World Model with Point Latents

Moaw: Unleashing Motion Awareness for Video Diffusion Models

NeXT-IMDL: Build Benchmark for NeXT-Generation Image Manipulation Detection & Localization

READ: End-to-End Autonomous Driving Made Safer with Efficient Reinforcement Learning

ManiCoG: Training-Free Improvement for GUI Grounding via Manipulation Chains

OGGSplat: Open-Vocabulary Gaussian Growing for Expanded Field-of-View

MoCtrl4D: Precise and Efficient Motion-Guided 4D Content Generation

End-to-End Unified Dense 3D Geometry and Motion Perception

EmbodiedSAM: Online Segment Any 3D Thing in Real Time

Point3R: Streaming 3D Reconstruction with Explicit Spatial Pointer Memory

QuadricFormer: Scene as Superquadrics for 3D Semantic Occupancy Prediction

Oryx MLLM: On-Demand Spatial-Temporal Understanding at Arbitrary Resolution

Efficient Sparse Single-stage 3D Visual Grounding with Text-guided Pruning

InstaRevive: One-Step Image Enhancement via Dynamic Score Matching

ThinkBot: Embodied Instruction Following with Thought Chain Reasoning

NGTTA: Non-parametric Geometry-driven Test Time Adaption for 3D Point Cloud Segmentation

V2M: Visual 2-Dimensional Mamba for Image Representation Learning

PixelGaussian: Generalizable 3D Gaussian Reconstruction from Arbitrary Views

OccSora: 4D Occupancy Generation Models as World Simulators for Autonomous Driving

GlobalMamba: Global Image Serialization for Vision Mamba

SlimLLaVA: Automatic Pruning for Large Vision-language Models

Embodied Instruction Following in Unknown Environments

From Forgery to Authenticity: Image Anti-Forensics via Reconstruction and Artefact Elimination