Jiawang Bian

Assistant Professor@NTU. Spatial Intelligence. Ex: Oxford, ByteDance Seed, Meta, Amazon, MBZUAI, Adelaide.

[Jiawang Bian @JiawangB]

🚀 Introducing our recent work, PixWorld, on scene reconstruction and generation.

🧩 It's a unified pixel-space diffusion model that directly optimizes into a scene-level 3D Gaussian Splatting (3DGS) representation via differentiable rendering — no VAE/RAE needed.

🎯 Flexible input: single image, multi-view images, or text — all unified in one framework, all outputting a full 3DGS scene.

⚡ Faster: 4-step distilled → ~0.6s/scene, up to ~1000× faster than diffusion world models.

📄 https://t.co/hfb0RUJQpS
🌐 https://t.co/zuIFtkGZ8Q
💻 https://t.co/ohDK5RfpXa
🤗 https://t.co/sBvsuFu40H

[Sensen Gao @gao_sensen40956]

PixWorld — Unifying 3D Scene Generation & Reconstruction in Pixel Space.

Prior work splits the two into different spaces — reconstruction in pixel space, generation in latent space. PixWorld unifies both in one pixel-space diffusion model, and instead of optimizing an intermediate latent it optimizes directly into a 3D representation via differentiable rendering (no VAE/RAE).

⚡ 4-step distilled → ~0.6 s/scene, up to ~1000× faster than diffusion world models.
📄 https://t.co/oFUqKacZj2
🌐 https://t.co/OaysMDJFTK
💻 https://t.co/rNsWwIa3JS
🤗 https://t.co/42tIq5mnkf

Paper & project page are live — feedback / RTs / ⭐ / upvotes welcome! 🙏

[Jiawang Bian @JiawangB]

🎉 Excited to officially join @NTUsg CCDS as Nanyang Assistant Professor & NRF AI Fellow!

Launching the **Spatial Intelligence Lab** at NTU — pushing frontiers in:
• 3D/4D Vision & World Models
• Vision-Language Models (VLMs)
• Vision-Language-Action (VLA)

Previously: Research Scientist @ByteDance Seed | Postdoc @Oxford & @MBZUAI

Open to collaborations, students, and conversations! 👇

https://t.co/ULhGC9TANC