Haotong Lin

A PhD student in State Key Laboratory of CAD & CG, Zhejiang University.

[Jiahui Lei @JiahuiLei1998]

The 2nd 4D Vision Workshop at CVPR is happening this Thursday afternoon in Room 506!

Come join us to see the latest progress in modeling the dynamic world. Looking forward to seeing you there! #CVPR2026 #CVPR

[Jianyuan@CVPR @jianyuan_wang]

Introducing VGGT-Ω: scaling feed-forward reconstruction across static and dynamic scenes, and studying whether the learned geometric representations transfer beyond reconstruction.

[Yuan Li @yuanli16342871]

Introducing: Long-Tail Internet Photo Reconstruction (CVPR’26)

We go beyond densely captured imagery to train more general 3D foundation models for the long tail of noisy, sparse, incomplete Internet photo collections of 3D scenes. Yet, we face a data bottleneck: models need ground truth for these long-tail scenes, which classical SfM fails to provide. How do we bypass it?

We break this bottleneck with two key contributions. First, we introduce MegaDepth-X, a large new dataset of scenes with high-quality 3D supervision. Second, we propose a new way to simulate difficult image sets for training.

Project page: https://t.co/Ti2dtTxjYG

[Ruojin Cai @ruojin8]

Excited to share our CVPR’26 work! Moving beyond dense captures to long-tail Internet photos, we introduce MegaDepth-X and sparsity-aware sampling for 3D reconstruction.

Great work led by Yuan (@yuanli16342871) and the team.

Project page: https://t.co/6lyuA1deHp

[Sida Peng @pengsida]

Happy to introduce Habitat-GS, a non-intrusive extension of Habitat-Sim that brings dynamic Gaussian Splatting for photorealistic rendering and comes with hundreds of high-quality 3DGS scene assets, aiming to empowering navigation research.
Code: https://t.co/2dyiRyCI7Y

[Haotong Lin @HaotongLin]

@jianyuan_wang The logo gives it away — 🥯mni 👀

[Haotong Lin @HaotongLin]

@CeyuanY Love this — one representation (text for cam pose), three roles, and the zero-motion generalization seals it 🙌 congrats again 🥯

[Haotong Lin @HaotongLin]

Congrats to the Omni team! 🥯🎉

Camera pose estimation? Just AR text gen 🤯
<campose>1.04 0.00 0.32 -0.20 -0.27 -0.02</campose>

6DoF floats as a string 🪢 no pose head, no bins, no 3D priors. On RealEstate10K: AUC@30 88.32 vs VGGT 88.23 on par with a geometry specialist!

[Ceyuan Yang @CeyuanY]

Introducing Omni, one unified model can support any-to-any multimodal modeling, including multimodal understanding, image/video generation and editing, world modeling and 3D reconstruction. All in one that adopts standard mixture-of-experts arch with only 3B activations.

[Haotong Lin @HaotongLin]

Amazing results!

[Haotong Lin @HaotongLin]

🍌perception is a subset of generation! Strong work — excited for what comes next on the unified vision front.

[Songyou Peng @songyoupeng]

Yay, finally! Introducing Vision Banana🍌 from @GoogleDeepMind, our unified model that outperforms SoTA specialist models on various vision tasks!

By treating 2D/3D vision tasks as image generation, we unlock a new foundation for CV.

Project page: https://t.co/GQgRi6mWwC

(1/5)

[Sida Peng @pengsida]

In our comparison on long image sequences, Scal3R consistently outperforms DA3.
In such cases, Scal3R offers a viable alternative.
Feel free to try Scal3R: https://t.co/SBVs4XQi2O

[NVIDIA AI Developer @NVIDIAAIDev]

Today, we released Lyra 2.0, a framework for generating persistent, explorable 3D worlds at scale, from NVIDIA Research.

Generating large-scale, complex environments is difficult for AI models. Current models often “forget” what spaces look like and lose track of movement over time, causing objects to shift, blur, or appear inconsistent. This prevents them from creating the reliable 3D environments required for downstream simulations. Lyra 2.0 solves these issues by:

✅ Maintaining per-frame 3D geometry to retrieve past frames and establish spatial correspondences
✅ Using self-augmented training to correct its own temporal drifting.

Lyra 2.0 turns an image into a 3D world you can walk through, look back, and drop a robot into for real-time rendering, simulation, and immersive applications.

➡️ Learn more: https://t.co/ROR7miJeCU
📄 Read the paper: https://t.co/1osU9EGjGD

[Haotong Lin @HaotongLin]

@NVIDIAAIDev Congrats on Lyra 2.0 👏 The video-side 3D consistency is the real star here — thrilled DA3 gets to ride along as the geometry backbone.

[Haian Jin@CVPR @Haian_Jin]

Spatial reconstruction is a long-context problem: real scenes come with hundreds of images. But O(N²) transformer-based models don’t scale efficiently.

Introducing: 🤐ZipMap (CVPR ’26): Linear-Time, Stateful 3D Reconstruction via Test-Time Training (TTT).

ZipMap “zips” a large image collection into an implicit TTT scene state in a single linear-time operation. The state will then be decoded into spatial outputs, and can be queried efficiently for novel-view geometry and appearance (~100 FPS)

ZipMap is not only much faster (>20× faster than VGGT), but also matches or surpasses the accuracy of all SOTA models.

[Saining Xie @sainingxie]

papers are kind of like movies: the first one is usually the best, and the sequels tend to get more complicated but not really more exciting. But that totally doesn’t apply to the DepthAnything series. @bingyikang's team somehow keeps making things simpler and more scalable each time.

in this new version, they basically show that a strong representation encoder plus a depth-ray prediction objective is enough (you see the RAE vibes too, right?) to get solid, general spatial perception across a bunch of tasks.

people often say they hate computer vision because it’s messy--too many tasks, too many data types, too many moving parts. but that’s exactly why I love it. I think the biggest AI breakthroughs are going to come quietly from vision and then suddenly leapfrog everything else, changing how AI interacts with the real world and with us.

pretty soon we’ll realize vision is not a big list of tasks--it’s a perspective. a perspective about modeling continuous sensory data, building layered representations of the world, and inching toward human-like intelligence. and tbh we’re watching this happen every day, behind all the hype, as all these different '"tasks" slowly start to merge.

[Bingyi Kang @bingyikang]

After a year of team work, we're thrilled to introduce Depth Anything 3 (DA3)! 🚀

Aiming for human-like spatial perception, DA3 extends monocular depth estimation to any-view scenarios, including single images, multi-view images, and video.

In pursuit of minimal modeling, DA3 reveals two key insights:
💎 A plain transformer (e.g., vanilla DINO) is enough. No specialized architecture.
✨ A single depth-ray representation is enough. No complex 3D tasks.

Three series of models have been released: the main DA3 series, a monocular metric estimation series, and a monocular depth estimation series.

The core team members, aside from me: @HaotongLin, Sili Chen, Jun Hao Liew, @donydchen.

👇(1/n)
#DepthAnything3

[Haotong Lin @HaotongLin]

@LordMassicotte @_akhaliq Not at the moment ⏳. Currently, Depth-Anything-3 does not support 360° camera inputs, but it's an interesting area for future exploration! 🔭

[Haotong Lin @HaotongLin]

@jianyuan_wang @bingyikang Thank you so much! We really appreciate you tweeting about our work!

[Haotong Lin @HaotongLin]

@Almorgand Not yet! Video depth support is coming soon 👀✨ Stay tuned — just a few more months!

[Haotong Lin @HaotongLin]

Thank you for sharing our work! Marigold is really cool! However, it’s somewhat limited by the image VAE — many flying points appear just after encoding a perfect ground-truth depth. Pixel-space diffusion to the rescue 🚀

[Anton Obukhov @AntonObukhov1]

Pixel-Perfect-Depth: the paper aims to fix Marigold's loss of sharpness induced by VAE by using VFMs (VGGT/DAv2) and a DiT-based pixel decoder to refine the predictions and achieve clean depth discontinuities. Video by authors.

[Haotong Lin @HaotongLin]

@sourav_bz @AntonObukhov1 Good question! Moge-v2 is a deterministic model — as discussed in our work, MSE loss makes it converge to middle depth near edges, causing flying points. You can see similar artifacts in Depth Anything V2, Moge V2, and Depth Pro— even though they produce very sharp 2D results.