Ziyu Wan

@raywzy1

Seattle, WA

raywzy.com

Joined August 2014

243Posts 249Followers 242Following

You might like

@JulienPhilip2

@LingjieLiu1

@GuyTvt

@SeonghyeonYe

@zhengqi_li

@david_roz_

@doubledaibo

@wenjing_bian

@_cheng_lin

@zhangboknight

@PhongStormVN

@zianwang97

@manuel_dahnert

@_ye_yuan

@YimingXie4

Our latest post explores on-policy distillation, a training approach that unites the error-correcting relevance of RL with the reward density of SFT. When training it for math reasoning and as an internal chat assistant, we find that on-policy distillation can outperform other…

thinkymachines's tweet image. Our latest post explores on-policy distillation, a training approach that unites the error-correcting relevance of RL with the reward density of SFT. When training it for math reasoning and as an internal chat assistant, we find that on-policy distillation can outperform other…

Ziyu Wan

@raywzy1

Oct 21

Want to know how to evaluate the spatial intelligence capability of your VLM? We will present VLM4D this afternoon! #ICCV2025

Shijie Zhou

@ShijieZhoucla

Oct 21

Can VLMs really think in 4D (3D space + time)? 🤔 When a model can’t tell “left” from “right,” something’s missing. That’s why we built VLM4D — a benchmark for spatiotemporal reasoning, debuting at #ICCV2025 📅 Oct 21 | 🕒 3–5 PM | 📍Exhibit Hall I #798 vlm4d.github.io

Ziyu Wan reposted

AK

@_akhaliq

Oct 17

Bee A High-Quality Corpus and Full-Stack Suite to Unlock Advanced Fully Open MLLMs

Ziyu Wan reposted

Guocheng Gordon Qian@ICCV

@guocheng_qian

Oct 15

I’ll be at ICCV! 🙋‍♂️Message me if you’re interested in joining us at Snap Research (Personalization Team) — we’re hiring research interns year-round in 🖼️Image editing 🎨Personalized generation 🤖Agentics 🧠VLMs for generation We also have full-time Research Scientist roles open!

guocheng_qian's tweet image. I’ll be at ICCV! 🙋‍♂️Message me if you’re interested in joining us at Snap Research (Personalization Team) — we’re hiring research interns year-round in
🖼️Image editing
🎨Personalized generation
🤖Agentics
🧠VLMs for generation
We also have full-time Research Scientist roles open!

Ziyu Wan

@raywzy1

Oct 1

👍👍👍

Shenao Zhang

@ShenaoZhang

Oct 1

🚀Excited to share our recent research:🚀 “Learning to Reason as Action Abstractions with Scalable Mid-Training RL” We theoretically study 𝙝𝙤𝙬 𝙢𝙞𝙙-𝙩𝙧𝙖𝙞𝙣𝙞𝙣𝙜 𝙨𝙝𝙖𝙥𝙚𝙨 𝙥𝙤𝙨𝙩-𝙩𝙧𝙖𝙞𝙣𝙞𝙣𝙜 𝙍𝙇. The findings lead to a scalable algorithm for learning action…

ShenaoZhang's tweet image. 🚀Excited to share our recent research:🚀

“Learning to Reason as Action Abstractions with Scalable Mid-Training RL”

We theoretically study 𝙝𝙤𝙬 𝙢𝙞𝙙-𝙩𝙧𝙖𝙞𝙣𝙞𝙣𝙜 𝙨𝙝𝙖𝙥𝙚𝙨 𝙥𝙤𝙨𝙩-𝙩𝙧𝙖𝙞𝙣𝙞𝙣𝙜 𝙍𝙇.
The findings lead to a scalable algorithm for learning action…

Ziyu Wan reposted

Microsoft Research

@MSFTResearch

Sep 10

RenderFormer, from Microsoft Research, is the first model to show that a neural network can learn a complete graphics rendering pipeline. It’s designed to support full-featured 3D rendering using only machine learning—no traditional graphics computation required. Learn more:…

Ziyu Wan reposted

Keenan Crane

@keenanisalive

Sep 6

“Everyone knows” what an autoencoder is… but there's an important complementary picture missing from most introductory material. In short: we emphasize how autoencoders are implemented—but not always what they represent (and some of the implications of that representation).🧵

keenanisalive's tweet image. “Everyone knows” what an autoencoder is… but there's an important complementary picture missing from most introductory material.

In short: we emphasize how autoencoders are implemented—but not always what they represent (and some of the implications of that representation).🧵

Ziyu Wan reposted

Wenzel Jakob {deprecation notice}

@wenzeljakob

Aug 7

Instead of blending colors along rays and supervising the resulting images, we project the training images into the scene to supervise the radiance field. Each point along a ray is treated as a surface candidate, independently optimized to match that ray's reference color.

wenzeljakob's tweet image. Instead of blending colors along rays and supervising the resulting images, we project the training images into the scene to supervise the radiance field.
Each point along a ray is treated as a surface candidate, independently optimized to match that ray's reference color.

Ziyu Wan

@raywzy1

Jul 15

simple but effective way to inject 3D info into your world simulator😉

Tianyu He

@deeptimhe

Jul 14

✅Videos are 2D projections of our dynamic 3D world. 🧠But can video diffusion models implicitly learn 3D information through training on raw video data, without explicit 3D supervision? ❌Our answer is NO !!!

deeptimhe's tweet image. ✅Videos are 2D projections of our dynamic 3D world.

🧠But can video diffusion models implicitly learn 3D information through training on raw video data, without explicit 3D supervision?

❌Our answer is NO !!!

Ziyu Wan reposted

International Conference on 3D Vision

@3DVconf

Jul 11

📢 The submission portal for #3DV2026 is LIVE on OpenReview 👉 openreview.net/group?id=3DV/2… Ready to ride the wave? 🌊 Deadline Aug 18 (≈ 5 weeks)—bring your best and make a splash!

3DVconf's tweet image. 📢 The submission portal for #3DV2026 is LIVE on OpenReview
👉 openreview.net/group?id=3DV/2…

Ready to ride the wave? 🌊
Deadline Aug 18 (≈ 5 weeks)—bring your best and make a splash!

Ziyu Wan reposted

Saining Xie

@sainingxie

Jun 30

awesome work by @jiacheng_chen_ and @sanghyunwoo1219 on 3D-grounded visual compositing (and nice demos!)

Sanghyun Woo

@sanghyunwoo1219

Jun 30

Introducing BlenderFusion: Reassemble your visual elements—objects, camera, and background—to compose a new visual narrative. Play the interactive demo: blenderfusion.github.io

Ziyu Wan

@raywzy1

Mar 21

🤯!!!

Shlomi Fruchter

@shlomifruchter

Dec 17

Let's test #Veo2 by dropping fruits into water! 🍓🍏🍎🫐 Blueberries first

Ziyu Wan

@raywzy1

Mar 3

“Introduction to Algorithms” is truly a phenomenal textbook! (Fun fact: back in high school, my laptop and this 1,300-page giant were inseparable desk buddies 😂😂😂)

raywzy1's tweet image. “Introduction to Algorithms” is truly a phenomenal textbook! (Fun fact: back in high school, my laptop and this 1,300-page giant were inseparable desk buddies 😂😂😂)

MIT CSAIL

@MIT_CSAIL

Mar 1

MIT’s “Introduction to Algorithms,” published #otd in 1990, is the world’s most cited CS text, with 67K citations & over a million copies sold. bit.ly/3y1yMPR @mitpress

MIT_CSAIL's tweet image. MIT’s “Introduction to Algorithms,” published #otd in 1990, is the world’s most cited CS text, with 67K citations &amp; over a million copies sold.

bit.ly/3y1yMPR

@mitpress

Ziyu Wan reposted

Jon Finger

@mrjonfinger

Feb 15

I grew up editing action movies with two vhs players & having to warn police we were filming with painted toy guns. My first VFX was frame by frame in photoshop from minidv. Kids growing up with advanced versions of this tech are going to do absolutely incredible things.

Ziyu Wan reposted

Kfir Aberman

@AbermanKfir

Feb 10

We discovered that imposing a spatio-temporal weight space via LoRAs on DIT-based video models unlocks powerful customization! It captures dynamic concepts with precision and even enables composition of multiple videos together!🎥✨

Ziyu Wan

@raywzy1

Feb 5

Inverse (direct G-buffer estimation) and forward rendering (no light transport simulation) using video diffusion model! Congratulations on the great work🥳🥳🥳

Zian Wang

@zianwang97

Jan 31

🚀 Introducing DiffusionRenderer, a neural rendering engine powered by video diffusion models. 🎥 Estimates high-quality geometry and materials from videos, synthesizes photorealistic light transport, enables relighting and material editing with realistic shadows and reflections