Qianyi Wu

Research Scientist @ Bytedance Seed, Prev: Ph.D @ Monash University

[Jiayi Weng @Trinkle23897]

Codex grew programmatic policies with no neural nets: max score on Breakout, and SOTA-level scores on MuJoCo.

Maybe heuristics were not too weak. Maybe they were just too expensive to maintain. Maybe it's the next paradigm.

https://t.co/1ZaIneleuW

[Chongjie Ye @ychngji6]

3D is code. Objects are the functions.
Pixels render worlds. Objects build them.

[Haven Feng @HavenFeng]

✨Thinking with Blender~
Meet VIGA: a multimodal agent that autonomously codes 3D/4D blender scenes from any image, with no human, no training!
@berkeley_ai #LLMs #Blender #Agent 🧵1/6

[Jianfeng Zhang @jianfengzhang95]

🚀 Thrilled to introduce Seed3D 1.0, a foundation model that generates High-Fidelity, Simulation-Ready 3D Assets directly from a Single Image!

✨ Key Capabilities:

1️⃣ High-fidelity Assets: Generates assets with accurate geometry, well-aligned textures, and physically-based materials.

2️⃣ Direct Integration: Assets integrate into physics engines with minimal configuration, enabling deployment in robotic manipulation and simulation training.

3️⃣ Scalable Generation: Beyond individual objects, scales to complete scene generation by assembling objects into coherent environments.

Seed3D 1.0 provides a foundation for advancing physics-based world simulators.

👇 Explore the full details, architecture, and results in the thread below.

📄 Paper: https://t.co/DKF2909bm4
🌐 Project: https://t.co/LoQl5eH9Yp

#3DGeneration #EmbodiedAI #WorldSimulators #FoundationModel

[Qianyi Wu @QianyiWu7]

A great opportunity if you are interested in conducting vision or AI research. Chuanxia is a great mentor you shouldn’t miss.

[Chuanxia Zheng @ChuanxiaZ]

After two amazing years with @Oxford_VGG, I will be joining @NTUsg as a Nanyang Assistant Professor in Fall 2025!

I’ll be leading the Physical Vision Group (https://t.co/byLxP7FE4a) — and we're hiring for next year!🚀

If you're passionate about vision or AI, get in touch!

[Qianyi Wu @QianyiWu7]

Wanna scale your feed-forward Gaussian Splatting model to the amazing 4K resolution?
Come check out our #CVPR2025 poster PanSplat at ExHall D, Poster #74 on Saturday, June 14, 10:30–12:30 and discuss with @chengzhag @haofeixu

[Cheng Zhang @chengzhag]

@QianyiWu7 @janusch_patas Excited to share that our PanSplat 🍳 poster will be at CVPR 2025, ExHall D, Poster #74 on Saturday, June 14, 10:30–12:30!
Stop by if you’re around — happy to chat and exchange ideas!
https://t.co/Xvc4ubjk1a

[Qianyi Wu @QianyiWu7]

Interesting work!

[Donny Y. Chen @donydchen]

ZPressor, an architecture-agnostic module that enables existing feed-forward #3DGS models to effectively handle dense input views (up to 100 on an 80G GPU). More at: https://t.co/HFzCPmeMk0

[Chong Zeng @iam_NCJ]

Big thanks to @_akhaliq for the shoutout!🥳

Our #SIGGRAPH2025 paper RenderFormer does end-to-end rendering from triangle meshes to images, with global illumination��� —— using just a simple transformer!!!🪄

🔗Project page: https://t.co/vk6cA4FgOE
💻Code: https://t.co/08Goe9mI13

[Lucas Beyer (bl16) @giffmana]

OK, ByteDance Seed is now firmly a top tier lab in my mind. Congrats on many solid works recently, continuously publishing and even releasing models.

A shame that I am really really bad at remembering individual Chinese names though :-/

[Tobias Fischer @TobiasFischer11]

1/3 Introducing FlowR 🌸: Flowing from Sparse to Dense 3D Reconstructions

We learn a direct mapping between incorrect renderings and their corresponding ground-truth images, augmenting scene captures with consistent novel, generated views to improve reconstruction quality.

[Anpei Chen @AnpeiC]

Why train when you can adapt? Easi3R unlocks training-free motion estimation from DUSt3R using attention adaptation—no fine-tuning needed! 💡

[Hao Zhao @HaoZhao_AIRSUN]

Powered by normals, driven by first principles — a long-term triumph of intrinsic decomposition in computer vision. We believe this is the ultimate answer to reconstructing the physical world.

[Siyuan Huang @siyuanhuang95]

🚀 Glad to share our new work accepted by CVPR2025! 🚀

Decompositional Neural Scene Reconstruction with Generative Diffusion Prior ��🎭

🔹 DP-Recon reconstructs high-quality interactive worlds from just 10 views!
🔹 Navigate, drag objects, edit geometry & texture via text ✏, and apply photorealistic VFX 🎬.
🔹 Handles large, heavily occluded scenes—outperforms baselines with 100 views using just 10!
🔹 Generalizes to in-the-wild 🌍 (e.g., YouTube videos) with detailed geometry & appearance from just 15 views.

📜 Read more: https://t.co/wAGLApRWi7
#NeRF #3DReconstruction #DiffusionPrior #SceneEditing

[Qianyi Wu @QianyiWu7]

@StanSzymanowicz Impressive results!

[AK @_akhaliq]

ReCamMaster

Camera-Controlled Generative Rendering from A Single Video

[Qianyi Wu @QianyiWu7]

@DinoFuzi Very impressive results!

[Radiance Fields @RadianceFields]

3DGS Compression is getting another big step up from HAC++: Towards 100X Compression of 3D Gaussian Splatting and has code already!

Project: https://t.co/4sm69cpIh3
Code: https://t.co/azet6NYOyJ

[Qianyi Wu @QianyiWu7]

Thx @janusch_patas for sharing our latest 🍳Pan-Series work (PanFusion https://t.co/FHYElhYVsX and then PanSplat https://t.co/SZXDPVzaP7). The code of PanSplat 🍳is released at https://t.co/hTVxINz6Gl. Welcome to check the details!

[MrNeRF @janusch_patas]

PanSplat: 4K Panorama Synthesis with Feed-Forward Gaussian Splatting

Pro-Tip: Try their interactive demo on YouTube. Links below!

Contributions:
• We present PanSplat, a feed-forward approach that efficiently generates high-quality novel views using a spherical 3D Gaussian pyramid tailored for panorama formats.

• Our pipeline features a hierarchical spherical cost volume and Gaussian heads with local operations, enabling a two-step deferred backpropagation that efficiently scales to higher resolutions.

• PanSplat achieves state-of-the-art results with superior image quality across synthetic and real-world datasets, offering up to 70× faster inference speed compared to the SOTA method [16]. By supporting 4K resolution, PanSplat is a promising solution for immersive VR applications.

[Dan Xu @danxuhk]

Thanks @janusch_patas for twittering our work. We propose a single-frame feedforward Gaussian Splatting framework based on cycle-consistency optimization for joint 3D scene reconstruction and generation.

[Chongjie Ye @ychngji6]

Thrilled to share: our GS to Mesh 2.0 has landed in Kiri Engine! 🎉
After a year of experimenting with pose estimation, GS optimization, and mesh generation, we've cracked it.
Check out the code on https://t.co/iSjP7oS92u. Full details dropping in late December!