Anpei Chen
@AnpeiC
Followers
716
Following
194
Media
10
Statuses
51
Group head @Inception3D Lab Assistant Professor @Westlake_Uni https://t.co/ZIIpOtFKvd
Hangzhou
Joined April 2021
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
πΊ Join us in Hawaii for Wild3D! We're hosting our 2nd Workshop on 3D Modeling, Reconstruction & Generation in the Wild! Dive into 3D + 4D topics, from real-world reconstruction to video generative models & dynamic scene modeling π #Wild3D #ICCV2025
5
5
45
π½πππ£π ππ£π πππ’π Being-in-the-world is the basic state of human existence. by Martin Heidegger ππͺπ’ππ£π―π Inferencing via One model, One stage; Training in One day using One GPU. https://t.co/iRWtURvrDf by Yue Chen @faneggchen
fanegg.github.io
Human3R: Everyone Everywhere All at Once
Real time online 3D reconstruction of 3D scene and humans represented with SMPL. https://t.co/SMsxP4iZhT I don't get tired of looking at these results
0
6
53
4/4 πPage: https://t.co/HmEvUwLIWb πPaper: https://t.co/cB79wRo2j0 π»Code: https://t.co/3dbZv9BnPN Big thanks to the amazing team! @RoverXingyu, @faneggchen, @yuliangxiu, Andreas Geiger, @AnpeiC
github.com
A simple state update rule to enhance length generalization for CUT3R - Inception3D/TTT3R
0
0
13
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.
1
2
14
#TTT3R: 3D Reconstruction as Test-Time Training We offer a simple state update rule to enhance length generalization for #CUT3R β No fine-tuning required! πPage: https://t.co/HmEvUwMgLJ 1/4 We rebuilt @taylorswift13βs "22" live at the 2013 Billboard Music Awardsβin 3D
5
39
289
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,
2
11
71
π» ππ
π½πΉπΌπΏπ² πΌππΏ π₯π²πππΉππ & ππΌπ±π² β’ Demos & videos: https://t.co/AGLXOYMDjT β’ Preprint on arXiv: https://t.co/SKLhGO9lZc
0
0
7
π’ Our new paper GaVS β 3D-Grounded Video Stabilization is out! Key idea: feed-forward Dynamic Gaussian Splatting + test-time optimization Robust, consistent, and cropping-free πΉ π₯ Project: https://t.co/88XWoJozKn
@youzn99 @stam_g @SiyuTang3 Dengxin Dai #SIGGRAPH25 #3DGS
4
25
198
Feature up up up πΌοΈβ¨ We tackle the resolution bottleneck of Vision Foundation Models (like DINOv2 & CLIP) with a coordinate-based cross-attention upsampler. Plug and play β stronger, faster than ever! π https://t.co/CrKBIiGlrT
#VisionModels #DeepLearning #ComputerVision
Introducing LoftUp! A stronger (than ever) and lightweight feature upsampler for vision encoders that can boost performance on dense prediction tasks by 20%β100%! Easy to plug into models like DINOv2, CLIP, SigLIP β simple design, big gains. Try it out! https://t.co/s09BLF8x1e
1
4
58
I love this new function! Never miss a beat again. https://t.co/ELeO8lVzof
0
0
4
Main contributions: π₯ Reconstruction-driven video diffusion model π Cyclical fusion of reconstruction and generation π New benchmark for NVS: Masked View Synthesis
0
0
1
Too many artifacts for GS reconstruction? Please checkout GenFusion: Closing the Loop between Reconstruction and Generation via Videos π Project page: https://t.co/z52j0IIsYU π» Code: https://t.co/UKHjnibZoE
#3D #DiffusionModels #ViewSynthesis #GenFusion #CVPR2025
3
39
197
Why train when you can adapt? Easi3R unlocks training-free motion estimation from DUSt3R using attention adaptationβno fine-tuning needed! π‘
π¦£Easi3R: 4D Reconstruction Without Training! Limited 4D datasets? Take it easy. #Easi3R adapts #DUSt3R for 4D reconstruction by disentangling and repurposing its attention maps β make 4D reconstruction easier than ever! πPage: https://t.co/9BngrGu7EL
1
17
99
How much 3D do visual foundation models (VFMs) know? Previous work requires 3D data for probing β expensive to collect! #Feat2GS @CVPR 2025 - our idea is to read out 3D Gaussains from VFMs features, thus probe 3D with novel view synthesis. πPage: https://t.co/ArpAbYKn33
4
33
260
How can we generate high-fidelity, complex 3D scenes? @QTDSMQ's LT3SD decomposes 3D scenes into latent tree representations, with diffusion on the latent trees enabling seamless infinite 3D scene synthesis! w/ @craigleili, @MattNiessner
https://t.co/wv9bIhkkYi
3
77
318
#3DV2025AMA very first guest, Michael J. Black from MPI-IS & Meshcapade @Michael_J_Black! π π You have now 24 HOURS to ask him anything β drop your questions in the comments below. Let's keep it respectful and engaging!
27
7
79
Struggling to secure more GPUs for training large X (reconstruction, Gaussian, etc) models? Check out LaRa, a lightweight 3D vision model designed to efficiently handle large-baseline reconstruction challenges https://t.co/nVSX2qx4ol
apchenstu.github.io
LaRa builds a feed-forward 360Β° bounded radiance field model in two days using 4 GPUs.
1
11
128