Group head @Inception3D Lab Assistant Professor @Westlake_Uni https://xitter.cf/t.co/ZIIpOtFKvd
Hangzhou
Joined April 2021
- Tweets67
- Following402
- Followers849
- Likes261
1/5 Excited to share an early exploration.
Coding agents like Codex can already tackle much of real-to-sim—even for robot manipulation. Does that mean 3D vision is solved?
Not quite. 3D still matters.
yubel426.github.io/Plumb/
4/5 One surprising takeaway: the harness can be remarkably lightweight.
Plumb is essentially a set of tools—without a hand-crafted workflow or rules telling the agent how to use them—and it already works well.
5/5 The question remains: what better tools should we build and put in agents’ hands next?
Page: yubel426.github.io/Plumb/
Code: github.com/Inception3D/Plumb
Big thanks to the amazing team!
@yubel426 @Hyuan_Chen @binbinhuang00
Anpei Chen retweeted
Excited to share our new work R³: 3D Reconstruction via Relative Regression.
Only 372M params (~⅓ of recent 1B-class baselines), trained on 6×48G GPUs, but competitive on streaming reconstruction. Runs at 30+ FPS.
Project: kevinxu02.github.io/r3-site/
Paper: arxiv.org/abs/2605.26519
We suggest going back to relative pose modeling,
enabling efficient and robust 3D reconstruction with low memory overhead — in both streaming and offline settings.
Project: kevinxu02.github.io/r3-site/
Paper: arxiv.org/abs/2605.26519
GlobalSplat: Stop unprojecting, start decoding. 🛠️
We fuse all input views into a fixed set of Global Scene Tokens to build high-fidelity 3D assets without the pixel-wise redundancy.
✅ Higher quality
✅ Better spatial allocation
🔗 r-itk.github.io/globalsplat/
#3DGS
#Motion324: 3D Motion Reconstruction for 4D Synthesis
We offer a feed-forward framework that synthesizes high-quality 4D assets from just a single monocular video.
 ✅ Mesh
 ✅ 3D Motion
 ✅ Feed-Forward
 ✅ Motion Retargeting
Check out 👇
motion3-to-4.github.io
Anpei Chen retweeted
If you are interested in 3D/4D/Video models, join us tomorrow (10/20) at the #ICCV #Wild3D workshop (Rm 312)!
We have an amazing set of all-star speakers! It will be fun! :)
@QianqianWang5
@AnpeiC
@Jimantha
Andrea Vedaldi
@angelaqdai
@JunGao33210520
@georgiagkioxari
𝘽𝙚𝙞𝙣𝙜 𝙖𝙣𝙙 𝙏𝙞𝙢𝙚
Being-in-the-world is the basic state of human existence.
by Martin Heidegger
𝙃𝙪𝙢𝙖𝙣𝟯𝙍
Inferencing via One model, One stage;
Training in One day using One GPU.
fanegg.github.io/Human3R/
by Yue Chen @faneggchen
Real time online 3D reconstruction of 3D scene and humans represented with SMPL.
fanegg.github.io/Human3R/
I don't get tired of looking at these results
#TTT3R: 3D Reconstruction as Test-Time Training
We offer a simple state update rule to enhance length generalization for #CUT3R — No fine-tuning required!
🔗Page: rover-xingyu.github.io/TTT3R
1/4
We rebuilt @taylorswift13’s "22" live at the 2013 Billboard Music Awards—in 3D
3/4
Instead of updating all states uniformly, we incorporate image attention as per-token learning rates.
High-confidence matches get larger updates, while low-quality updates are suppressed.
4/4
🔗Page: rover-xingyu.github.io/TTT3R
📄Paper: arxiv.org/abs/2509.26645
💻Code: github.com/Inception3D/TTT3R
Big thanks to the amazing team!
@RoverXingyu, @faneggchen, @yuliangxiu, Andreas Geiger, @AnpeiC
The fields are moving extremely fast, we tried to summarize them base on 3D representations.
Please let us know if we missed anything :)
Advances in Feed-Forward 3D Reconstruction and View Synthesis: A Survey
Jiahui Zhang, Yuelei Li, @AnpeiC, Muyu Xu, Kunhao Liu, @jianyuan_wang, @xxlong0, @hx_liang95, @zexiangxu, @haosu_twitr, Christian Theobalt, Christian Rupprecht, Andrea Vedaldi, @hpfister, Shijian Lu, @fnzhan0507
tl;dr: in title
arxiv.org/abs/2507.14501
📢 Our new paper GaVS – 3D-Grounded Video Stabilization is out!
Key idea: feed-forward Dynamic Gaussian Splatting + test-time optimization
Robust, consistent, and cropping-free 📹
🎥 Project: sinoyou.github.io/gavs
@youzn99 @stam_g @SiyuTang3 Dengxin Dai
#SIGGRAPH25 #3DGS
💻 𝗘𝘅𝗽𝗹𝗼𝗿𝗲 𝗼𝘂𝗿 𝗥𝗲𝘀𝘂𝗹𝘁𝘀 & 𝗖𝗼𝗱𝗲
• Demos & videos: lnkd.in/eDPcx6Wv
• Preprint on arXiv: lnkd.in/ex6EG3D5