Minghan Qin

Researcher@Bytedance Seed. Focusing on 3D, Agent, VLM, and world model.

[Minghan Qin @qin_minghan]

@wildmindai Thanks so much for sharing our work :)

[Wildminder @wildmindai]

Lucida from ByteDance

Cool project! It transforms indoor video into editable 3D scenes; reconstructs objects as simulation-ready assets
- VLM-based object detection and referring descriptions.
- good at pose alignment
- high-fidelity real-to-sim digital twins.
- multi-view masks, boxes, partial point clouds
- Seed3D 2.0 for image-to-3D asset gen

https://t.co/uUlcInA2XL

[Minghan Qin @qin_minghan]

🚀 Introducing Lucida: composable Real-to-Sim scene modeling—turning real-world captures into composable 3D worlds.
Project Page: https://t.co/incxYK0EBG
HF: https://t.co/JNUKRJW7gU
Arxiv: https://t.co/mVgF6TQPpI

#Real2Sim #3DVision #EmbodiedAI #Robotics #ComputerVision #3DGen

[Minghan Qin @qin_minghan]

Lucida: Parse, Generate, and Place for Composable Real-to-Sim Scene Mode... https://t.co/kCz9dB3Ytu via @YouTube

[Minghan Qin @qin_minghan]

Thanks for sharing our work! @HuggingPapers
We are excited to announce that SIMART has been accepted to #SIGGRAPH2026!

[DailyPapers @HuggingPapers]

SIMART: Decomposing monolithic meshes into sim-ready articulated assets

A unified MLLM framework using Sparse 3D VQ-VAE reduces tokens by 70% to enable part-level decomposition and kinematic prediction for physics-based robotic simulation.

[Minghan Qin @qin_minghan]

Thrilled to share that our paper SIMART has been accepted to #SIGGRAPH2026! 🚀
Check out the full demo and details:
Project Page: https://t.co/cQ47aHYVU9
ArXiv: https://t.co/IiFfCWdcQh
Demo Video: https://t.co/fHED9yF1uL
#EmbodiedAI #MLLM #SIGGRAPH2026 #AI

[Minghan Qin @qin_minghan]

SIMART: Decomposing Monolithic Meshes into Sim-ready Articulated Assets ... https://t.co/6GJRiuBBbQ via @YouTube

[Minghan Qin @qin_minghan]

Thanks to @_akhaliq for sharing our work on #MeshCoder !
Project Page: https://t.co/aJfFQEy3mb
Paper: https://t.co/THLCu6PZmk
Code: https://t.co/xyK6g0byua

[AK @_akhaliq]

MeshCoder

LLM-Powered Structured Mesh Code Generation from Point Clouds

[Minghan Qin @qin_minghan]

Thanks @janusch_patas for sharing our work!

[MrNeRF @janusch_patas]

GUAVA: Generalizable Upper Body 3D Gaussian Avatar

Contributions:
• We propose GUAVA, the first framework for generalizable upper-body 3D Gaussian avatar reconstruction from a single image. Using projection sampling and inverse texture mapping, GUAVA enables fast feed-forward inference to reconstruct Ubody Gaussians from the image.

• We introduce an expressive human template model with a corresponding upper-body tracking framework, providing an accurate prior for reconstruction.

• Extensive experiments show that GUAVA outperforms existing methods in rendering quality and significantly outperforms 2D diffusion-based methods in speed, offering fast reconstruction and real-time animation.

[MrNeRF @janusch_patas]

4D LangSplat: 4D Language Gaussian Splatting via Multimodal Large Language Models

Contributions:
• We introduce 4D LangSplat for open-vocabulary 4D spatial-temporal queries. To the best of our knowledge, we are the first to construct 4D language fields with object textual captions generated by MLLMs.

• To model smooth transitions across states for objects in 4D scenes, we propose a status deformable network to capture continuous temporal changes.

• Experiential results show that our method attains state-of-the-art performance for both time-agnostic and time-sensitive open-vocabulary queries.

[MrNeRF @janusch_patas]

"LangSplat: 3D Language Gaussian Splatting." Everything including code released but the paper is not out yet 🎅
https://t.co/ohnIuZKFif
https://t.co/Ssakr5sL41

[Minghan Qin @qin_minghan]

Thanks to @_akhaliq for sharing our work #LangSplatV2!

Website: https://t.co/QJKdMFjeEA
Paper: https://t.co/BlbFHRAc0V
Code:https://t.co/x7kodFQpLH

[AK @_akhaliq]

LangSplatV2

High-dimensional 3D Language Gaussian Splatting with 450+ FPS

[Minghan Qin @qin_minghan]

Thanks to @_akhaliq for sharing our #CVPR2025 work on #4DLangSplat!

Project page: https://t.co/96WJ7nXJXg
paper: https://t.co/gCwtiWLlJJ
Code: https://t.co/KrTr4Rg4Bc

[AK @_akhaliq]

4D LangSplat

4D Language Gaussian Splatting via Multimodal Large Language Models

[Minghan Qin @qin_minghan]

Thanks @_akhaliq for sharing our work!
project page: https://t.co/iYECUwObAF
paper: https://t.co/EVHD8cVoLv
Code: https://t.co/QD5RlvZZnQ

[AK @_akhaliq]

LangSplat: 3D Language Gaussian Splatting

paper page: https://t.co/RtOpGlO8lS

ground CLIP features into a set of 3D language Gaussians, which attains precise 3D language fields while being 199 × faster than LERF

[Tsinghua University @Tsinghua_Uni]

Tsinghua and @Harvardresearchers introduced #LangSplat, an advanced #AI method for 3D language fields. It uses 3D Gaussian Splatting to enhance user communication with #3Denvironments.
For more: https://t.co/HY591hSs0M