Chong Zeng

CS PhD Student @Stanford | Previous: @ZJU_china @MSFTResearch @UCSanDiego

[Chong Zeng @iam_NCJ]

What if a Transformer could render?
Not text → image.
But mesh → image — with global illumination.

No rasterizers. No ray-tracers. Just a Transformer without per-scene training.

RenderFormer does exactly that.

#SIGGRAPH2025
🔗https://t.co/vk6cA4FgOE

[Chong Zeng @iam_NCJ]

@sunfanyun @moonlake Just to clarify — these are EXACTLY from our RenderFormer paper — both the figures and the core idea. (https://t.co/vk6cA4FOEc)
No collab or connection with Moonlake.
Not cool to use them for startup promo without asking.
cc @MSFTResearch

[Chong Zeng @iam_NCJ]

Thanks for all the participants! Our RenderFormer has been selected top-10 best fast forward presentations in #SIGGRAPH2025 🎉🎉🎉

[ACM SIGGRAPH @siggraph]

🏆During the Technical Papers Fast Forward, participants voted for their favorite presentations — and here are the top 10 picks that captured the most attention, creativity, and innovation this year!

🖌 Painting with 3D Gaussian Splat Brushes

🍽 VirCHEW Reality: On-Face Kinesthetic Feedback for Enhancing Food-Intake Experience in Virtual Reality.

✨ SpotLessSplats: Ignoring Distractors in 3D Gaussian Splatting

💡 RenderFormer: Transformer-based Neural Rendering of Triangle Meshes with Global Illumination

🌌 Don’t Splat your Gaussians: Volumetric Ray-Traced Primitives for Modeling and Rendering Scattering and Emissive Media

💧 Gaussian Fluids: A Grid-Free Fluid Solver based on Gaussian Spatial Representation

👐 VR-Doh: Hands-on 3D Modeling in Virtual Reality

🌟 What is HDR? Perceptual Impact of Luminance and Contrast in Immersive Displays

👀 Towards Understanding Depth Perception in Foveated Rendering

💥 BANG: Dividing 3D Assets via Generative Exploded Dynamics

👏 Thank you to everyone who voted and celebrated these incredible achievements!

photo by John Fujii © 2025 ACM SIGGRAPH

[Chong Zeng @iam_NCJ]

@AntonObukhov1 @_akhaliq Good suggestion — sounds feasible, we’ll give it a try!

[Chong Zeng @iam_NCJ]

Curious how far a pure Transformer can go in rendering — and in building broader world models?

Check out the paper, code, and more:
🔗 https://t.co/vk6cA4FgOE
📄 https://t.co/xmzUKekQEB
💻 https://t.co/08Goe9mI13

[Chong Zeng @iam_NCJ]

What if a Transformer could render?
Not text → image.
But mesh → image — with global illumination.

No rasterizers. No ray-tracers. Just a Transformer without per-scene training.

RenderFormer does exactly that.

#SIGGRAPH2025
🔗https://t.co/vk6cA4FgOE

[Chong Zeng @iam_NCJ]

It also works with 𝐩𝐡𝐲𝐬𝐢𝐜𝐚𝐥-𝐛𝐚𝐬𝐞𝐝 𝐬𝐢𝐦𝐮𝐥𝐚𝐭𝐢𝐨𝐧𝐬 —
If it’s a triangle mesh, RenderFormer can render it.

[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

[AK @_akhaliq]

Microsoft releases RenderFormer on Hugging Face

Transformer-based Neural Rendering of Triangle Meshes with Global Illumination

[Xiang Feng @f1she1]

🎉Excited to share that ARM: Appearance Reconstruction Model for Relightable 3D Generation, has been accepted at CVPR (✨Highlight)! ARM enables 3D generation with PBR materials for relighting under novel views/lighting. Check https://t.co/yeerJwxdJx for more results and details!

[Whispers from the Star @WFTS_Game]

Whispers from the Star⭐️ Announce Trailer

Your words seal her fate.

When a girl named Stella crash-lands on an alien planet called Gaia, you are the only person she can contact through her communicator. Through texts, voice messages, and video calls that unfold throughout your day, guide her through a gripping story where every response could mean life or death.

As you stay connected in real time, your open-ended conversations with Stella become her compass through danger. In this innovative story of hope, resilience, and wonder, you are Stella's sole chance at survival. Will your voice guide her home?

–––––––
➤ Register now for the Closed Beta:
https://t.co/mt8Pf99Gco
➤ Join the Discord: https://t.co/gqsd7983Mg

#WhispersfromtheStar #WFTS #Anuttacon

[Chong Zeng @iam_NCJ]

Cool blog and survey from @PulkitRega ! Let’s make lighting💡great again🎉🥳

[darthgera123 @PulkitRega]

Wrote a blog about how diffusion models are being used to solve relighting tasks such as scene editing, harmonization, etc. Included papers from @iam_NCJ @mengweir @zhanghesprinter @MattNiessner @jflalondeqc @GinterYohan @Jimantha @Haian_Jin @Peter4AI @JulienPhilip2.

[MrNeRF @janusch_patas]

Code release entered the room!

[SIGGRAPH Asia '24] GS^3: Efficient Relighting with Triple Gaussian Splatting

Link in the comments.

[Chong Zeng @iam_NCJ]

Happy to announce that MeshFormer is now accepted to #NeurIPS !🎉

[Minghua Liu @MinghuaLiu_]

Want high-quality 3D meshes with sharp geometric details? Try our newly released MeshFormer!

It only takes 8 GPUs for two days of training, outperforming state-of-the-art models that use over a hundred GPUs!

With 3D-native input guidance, representations, supervision, and post-processing, we significantly improve the training efficiency and geometric quality of feed-forward reconstruction models!

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

@iam_NCJ

[MrNeRF @janusch_patas]

GS^3: Efficient Relighting with Triple Gaussian Splatting

Abstract:
We present a spatial and angular Gaussian based representation and a triple splatting process, for real-time, high-quality novel lighting-and-view synthesis from multi-view point-lit input images.

To describe complex ap pearance, we employ a Lambertian plus a mixture of angular Gaussians as an effective reflectance function for each spatial Gaussian.

To generate self-shadow, we splat all spatial Gaussians towards the light source to obtain shadow values, which are further refined by a small multi-layer perceptron.

To compensate for other effects like global illumination, another network is trained to compute and add a per-spatial-Gaussian RGB tuple.

The effectiveness of our representation is demonstrated on 30 samples with a wide variation in geometry (from solid to fluffy) and appearance (from translucent to anisotropic), as well as using different forms of input data, including rendered images of synthetic/reconstructed objects, photographs captured with a handheld camera and a flash, or from a professional lightstage.

We achieve a training time of 40-70 minutes and a rendering speed of 90 fps on a single commodity GPU. Our results compare favorably with state-of-the-art techniques in terms of quality/performance.

[Chong Zeng @iam_NCJ]

Thanks @_akhaliq !
Check our project page https://t.co/BCeGYrqByv for more results & demo!🥂

[AK @_akhaliq]

MeshFormer

High-Quality Mesh Generation with 3D-Guided Reconstruction Model

discuss: https://t.co/RjYIt5aw3T

demo: https://t.co/cbUrvKywog

MeshFormer reconstructs high-quality 3D textured meshes with fine-grained, sharp geometric details in a single feed-forward pass that takes just a few seconds. MeshFormer can be trained using 8 H100 GPUs for just 2 days, whereas concurrent works require more than one hundred.

[Minghua Liu @MinghuaLiu_]

Want high-quality 3D meshes with sharp geometric details? Try our newly released MeshFormer!

It only takes 8 GPUs for two days of training, outperforming state-of-the-art models that use over a hundred GPUs!

With 3D-native input guidance, representations, supervision, and post-processing, we significantly improve the training efficiency and geometric quality of feed-forward reconstruction models!

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

@iam_NCJ

[Chong Zeng @iam_NCJ]

Just spent a happy week at Denver #SIGGRAPH2024 !
Looking forward to next year’s event~