Lu Sheng

I am a Professor at Beihang University, focusing on 3D computer vision. I am the PI of Kaleidoscope Laboratory at Beihang University.

[Zehuan-Huang @huanngzh]

⚽️ Model layout in 3D space

LaviGen #cvpr26, a native 3D layout generation model that models layout arrangement combining LLM knowledge with 3D generation priors.

Paper: https://t.co/K2jmvTiSnP
Project: https://t.co/Z4WLQLCl3C
Code (coming later): https://t.co/SRU3IgoNCy

[Zehuan-Huang @huanngzh]

Thanks for developing our SeviGen into @ComfyUI !
SegviGen, the SOTA 3D part segmentation method, has been conditionally accepted by SIGGRAPH 2026 (Journal Track). Feel free to try it!

Project: https://t.co/RrTBjWcrdj
Code: https://t.co/RGghaWcSr9

[Zehuan-Huang @huanngzh]

Thanks @_akhaliq for sharing our work. We have released our code. Feel free to try our #VoxHammer! Our online demo is coming soon. Stay tuned!

- Paper: https://t.co/8ygB8bUIg1
- Code: https://t.co/6WeLYH5JC8

[Zehuan-Huang @huanngzh]

Excited to share that MV-Adapter, a multi-view adapter on SD/SDXL series, has been accepted by
@ICCVConference !
Check our code in https://t.co/id78PbDg8T
and try our online demos in https://t.co/sRXGeIaAh9!

[AK @_akhaliq]

AnimaX

Animating the Inanimate in 3D with Joint Video-Pose Diffusion Models

[Zehuan-Huang @huanngzh]

MIDI-3D updates! An open-source full pipeline for single image to compositional **textured** 3d scene. We integrate MIDI-3D with #MVAdapter in a @Gradio demo. Check our source code: https://t.co/8O9LeNq3C0

[Unitree @UnitreeRobotics]

Unitree B2-W Talent Awakening! 🥳
One year after mass production kicked off, Unitree’s B2-W Industrial Wheel has been upgraded with more exciting capabilities.
Please always use robots safely and friendly.
#Unitree #Quadruped #Robotdog #Parkour #EmbodiedAI #IndustrialRobot #InspectionRobot #IntelligentRobot #FoundationModels #LeggedRobot #WheeledLegs

[Lu Sheng @SHENGLui1989]

MV-Adapter - Another impressive work done by Zehuan @huanngzh 👍👍👍. Adaptability, versatility, efficiency and high-resolution all in one🚀 for multi-view generation from text, image, geo. and etc.

Please try our demos 🔗 https://t.co/R7Gp4xt3rk

[Zehuan-Huang @huanngzh]

🔥Multi-view Generation Made Easy Now🔥

Excited to present MV-Adapter, a creative productivity tool that seamlessly transfer text-to-image models to multi-view generators.

[Code and demo released]
Project page: https://t.co/ieZ4Ux29Vt
Paper: https://t.co/a3BKuGDTuf

Highlight>>

[Zehuan-Huang @huanngzh]

🔥Multi-view Generation Made Easy Now🔥

Excited to present MV-Adapter, a creative productivity tool that seamlessly transfer text-to-image models to multi-view generators.

[Code and demo released]
Project page: https://t.co/ieZ4Ux29Vt
Paper: https://t.co/a3BKuGDTuf

Highlight>>

[Lu Sheng @SHENGLui1989]

I am proud to share our recent project #MIDI. Another impressive work done by Zehuan @huanngzh🚀

[Zehuan-Huang @huanngzh]

🚀Turn Single Image into 3D Scene🚀

#MIDI is a generalizable multi-instance diffusion model that turns single images into high-quality compositional 3D scenes.

Check the interactive results in our 🔗Project page: https://t.co/3P5ezmoXth.

📖Paper: https://t.co/BWtmD9xWI6

[Zehuan-Huang @huanngzh]

🚀Turn Single Image into 3D Scene🚀

#MIDI is a generalizable multi-instance diffusion model that turns single images into high-quality compositional 3D scenes.

Check the interactive results in our 🔗Project page: https://t.co/3P5ezmoXth.

📖Paper: https://t.co/BWtmD9xWI6

[Lu Sheng @SHENGLui1989]

@yanpei_cao @andy_yx27 Big congrats!㊗️

[Lu Sheng @SHENGLui1989]

@_akhaliq 🤩Thanks for sharing our work!
Project page: https://t.co/Hml4dqrq77
Paper: https://t.co/QXH4KC26YY
Code(coming soon): https://t.co/fjumIbKswZ

[AK @_akhaliq]

Ouroboros3D

Image-to-3D Generation via 3D-aware Recursive Diffusion

Existing single image-to-3D creation methods typically involve a two-stage process, first generating multi-view images, and then using these images for 3D reconstruction. However,

[Lu Sheng @SHENGLui1989]

🌟Catch our poster on Octavius this Friday afternoon (Hall B, #91) , a.k.a. one of the "first" LoRA-MoE papers about MLLMs😂 #ICLR2024

[Zehuan-Huang @huanngzh]

Introduce our work: 🦸‍♂️Parts2Whole: A Unified Reference Framework for Controllable Human Image Generation.
Parts2Whole generates human images in various postures from referential human part images of any quantity and different origins.
https://t.co/GTKPZNKdwZ

[Siyuan Huang @siyuanhuang95]

🔥🔥 Excited to announce that we are hosting the ARNOLD challenge at the Embodied AI Workshop this year!

It is the only challenge using the Issac Sim simulator.

Submit your results before June 2nd at https://t.co/0yH4N4Mr5r!

[Zhenfei (Jeremy) Yin @9LdROhjZE56jSh9]

Happy to share that our paper "MP5: A Multi-modal Open-ended Embodied System in Minecraft via Active Perception" was accepted at #CVPR2024 🎉

We have made an attempt in the area of multimodal embodied #agent systems on #Minecraft. By utilizing an active perception scheme based on #MLLM (Multimodal Large Language Model) and a #multiagent collaboration mechanism, we enable agents to solve tasks in open-ended visual environments that are not only long-horizon but require the perception of complex environmental information as well, such as "collecting sand ⛱️ at the bottom of the water 🌊 during the daytime ☀️ with a small wooden shovel."

Project Page: https://t.co/JablbNfg5z
LAMM Comunity: https://t.co/DECkHvVlbV

[Michael Black @Michael_J_Black]

OSSO data is finally online. Published at CVPR'22, OSSO infers the 3D skeleton inside #SMPL. The delay was due to the UKBiobank process. Because the source data came from them, they have to distribute our data as a "return" to the Biobank. Here's the link https://t.co/jVcwm9nGWS