Zhengzhong Tu @_vztu Twitter profile

Last Seen Profiles

@presidencymv

@Xts8KlrjSbQRnRK

@Bastos_Quotes

@Partnerdikamar

@cvptedbxtch

@Qnasalhroff90

@ESokullu

@OMC_mindfulness

@animalpolitico

@chelseabyrne003

@ApexAABruins

@thebritbar

@SingleyStuCo

@BETMskykid

@tsukiodysey

@DjDashGermany

@AndradeArj63995

@jauzy19

@Maran_arhab

@Olumoore13

@JayaHoIndia

@DerrickWalkerJ2

@_LarsPetersen

@NIPCS_ES

@SamueILFC

@mellissameisels

@Lukaszpolska

@mah2020xx

@stw_pdg

@AlnaouqA

@NYCsidewalker

@stw_pdg

@anonymousslyy_

@AnniceShal40899

@nationalrailenq

@EsmeBalles93535

Zhengzhong Tu

@_vztu

7 months

Computer vision has been solved.

181

3K

13K

Zhengzhong Tu

@_vztu

5 months

Reinforcement learning explained

156

2K

13K

Zhengzhong Tu

@_vztu

2 years

State-of-the-art human pose estimation model

20

574

3K

Zhengzhong Tu

@_vztu

2 years

Senior Engineer walking you through the production code

5

46

512

Zhengzhong Tu

@_vztu

2 years

"What a week, huh?" "Captain, it's Monday morning!" --- Happy to share that we've open-sourced the model weights and Colab demo of our #ECCV2022 paper "MaxViT: Multi-Axis Vision Transformer", finally! 🤗Pls check it out and play around with it:

10

63

433

Zhengzhong Tu

@_vztu

2 years

📢📢Check out our recent efforts that improve on the Vision Transformer and MLP models for hi-res inputs. Covers two papers from @GoogleAI - MAXIM ( #CVPR2022 ) and MaxViT ( #ECCV2022 ). Achieves SoTA results on more than ten high- and low-level visual tasks.

A Multi-Axis Approach for Vision Transformer and MLP Models

Posted by Zhengzhong Tu and Yinxiao Li, Software Engineers, Google Research Convolutional neural networks have been the dominant machine learning a...

research.google

Google AI

@GoogleAI

2 years

Computer vision models often struggle to handle high-res inputs. Here we describe a multi-axis approach that better accommodates such inputs to achieve excellent performance on both high- and low-level vision tasks, from object detection to deblurring. →

12

66

260

1

17

80

Zhengzhong Tu

@_vztu

2 years

Successfully defended my #PhD thesis today! Officially Dr. Tu now! 🥳🎓🤩 It’s time to say bye-bye to the greatest advisor @AlanBovik ; meanwhile more than happy to share that I will be joining @docmilanfar ’s team at Google Research in August!

8

1

64

Zhengzhong Tu

@_vztu

2 years

@lidangzzz 我也生活在上海，怎么我就感受不到高物价呢？宁没资格对咱们国家说三道四🥵

9

0

63

Zhengzhong Tu

@_vztu

2 years

Already back in Austin now! Finally caught up with @ak92501 the Arxiv robot on the last day of CVPR~

0

3

60

Zhengzhong Tu

@_vztu

6 months

Passionate about Computational Photography and Generative Models? Our team is hiring PhD student researchers & interns! DM me or @2ptmvd & apply here: 1. PhD Student Researchers (graduating after Dec 2025) Student Researcher, PhD, Winter/Summer 2024 — Google Careers 2. PhD…

mauricio delbracio

@2ptmvd

6 months

Passionate about Computational Photography and Generative Models? Our team at Google Research is hiring interns! DM me & apply: 1. PhD Student Researchers (graduating after Dec 2025) 2. PhD Research Interns (graduating by Dec 2025)

6

25

120

3

7

55

Zhengzhong Tu

@_vztu

2 years

Folks, while you're having fun at #ECCV2022 in Tel-Aviv, why not have a rest and check out our (virtual) poster here?👇😉 Paper: Video: Code is available: * Tensorflow: * PyTorch:

3

10

55

Zhengzhong Tu

@_vztu

5 months

Wowwwww. #NeurIPS2023

6

0

51

Zhengzhong Tu

@_vztu

10 months

What can you do with just 4️⃣ parameters? —- We present a surprisingly simple idea & method, and it works extremely well for broad vision tasks. 👉Our #ICCV2023 paper is perfect for a relaxed weekend read 📑 MULLER: Multilayer Laplacian Resizer for Vision

MULLER: Multilayer Laplacian Resizer for Vision

Image resizing operation is a fundamental preprocessing module in modern computer vision. Throughout the deep learning revolution, researchers have overlooked the potential of alternative resizing...

arxiv.org

Peyman Milanfar

@docmilanfar

10 months

MULLER: Multilayer Laplacian Resizer for Vision Super lightweight resizer w/ literally a handful of trainable parameters. Plugged into training pipelines, it significantly boosts the performance of the underlying vision task at ~no cost #ICCV2023 (1/3)

5

21

135

1

8

42

Zhengzhong Tu

@_vztu

5 months

1/🧵 Introducing "💲𝐓𝐈𝐏: Text-Driven Image Processing with Semantic and Restoration Instructions". 𝐓𝐈𝐏 allows users to flexibly play with 𝙿𝚛𝚘𝚖𝚙𝚝𝚜 (eg, "zebra", "orange", or "deblur", "upsample"). Arxiv: Project:

2

8

36

Zhengzhong Tu

@_vztu

2 years

Really excited that our paper "MAXIM: Multi-Axis MLP for Image Processing" made it to 1 of the 33 best paper finalists at #CVPR2022 ! Please come by our oral & poster tomorrow (6/22): 830-10am Oral 2.1.3 Low-level Vision (Great Hall B-C) 10-1230am Poster 2.1 #51 (Great Hall B2-C)

Peyman Milanfar

@docmilanfar

2 years

Happy to tell you that our MAXIM paper was a best paper finalist (1 of 33) at #CVPR2022 ! Kudos to @zhengzhongtu and his internship mentor @yinxiaoli for this nice work Paper: Supplementary Material: Code:

6

14

122

4

2

34

Zhengzhong Tu

@_vztu

2 years

📢 #ECCV2022 alert - we built two vision Transformers, 2/2 accepted to @eccvconf , both open-sourced!! (1/2): "MaxViT: Multi-Axis Vision Transformer" (2/2): "V2X-ViT: Vehicle-to-Everything Cooperative Perception with Vision Transformer" Check more details (paper&code) below 1/n

5

34

Zhengzhong Tu

@_vztu

5 months

GPT, we’ve had a problem. We have had one agent out of control.

1

3

31

Zhengzhong Tu

@_vztu

5 months

I will be ✈️ to #NeurIPS2023 at #NewOrleans 🎷 from 12/10-12/17. Excited to meet old friends again and say hello to new friends ☕️. Let’s chat about anything 🔥!

3

0

32

Zhengzhong Tu

@_vztu

2 years

PhD student when their advisor is traveling to attend a conference

0

26

Zhengzhong Tu

@_vztu

5 months

This new, more powerful version of Pixel 8 Photo Unblur feature is the first project I delivered after I joined #GoogleAI . It would have never happened without the amazing teamwork. I learned way more than I expected in bridging research and productization, and finally delivered…

Peyman Milanfar

@docmilanfar

5 months

This week, we shipped a new version of Photo Unblur on Pixel 8 that's even better, with a wider range of things it can fix. Examples in thread Kudos to the awesome team including @2ptmvd , @_vztu , @jia_xuhui , @YandongLi8 , @ErikaSporkert @navinsarmaphoto , @sebarod & many others 1/5

1

5

81

5

2

26

Zhengzhong Tu

@_vztu

3 months

You probably don’t know how complex Chinese chars are. Even me, a native speaker, under-estimate it. Today is the new year of Dragon (“龍”), and we have a greeting idiom: 龍行龘龘 (meaning “Soar the sky like dragon”, indicating being cheerful and upbeat about future)

3

0

26

Zhengzhong Tu

@_vztu

5 months

Wowwww again #NeurIPS2023

4

0

25

Zhengzhong Tu

@_vztu

6 months

Fresh PhD students passionate about AI research

0

24

Zhengzhong Tu

@_vztu

1 year

It's been a while, but I'm glad to share that MaxViT has been officially released through TF Model Garden! Supports training for multiple tasks -- imgnet classification pretrain/finetune, object detection (Mask RCNN, Retinanet), and semantic segmentation.

1

2

22

Zhengzhong Tu

@_vztu

6 months

#CVPR2024

@CVPR

6 months

What are you doing to unwind after the #CVPR2024 main paper deadline? Share a pic!

24

4

49

1

0

23

Zhengzhong Tu

@_vztu

5 months

Paper’s results vs your own implementation

0

21

Zhengzhong Tu

@_vztu

2 years

Finally, attaching some of the amazing restoration results from MAXIM👇. For more details, please check: Arxiv: code&model: Amazing collaborators @docmilanfar , @AlanBovik , @yinxiaoli , et al. Empowered by @GoogleAI & @UTAustin . n/n

0

7

22

Zhengzhong Tu

@_vztu

3 months

how it started how it ended

3

2

22

Zhengzhong Tu

@_vztu

2 years

@docmilanfar Escaped to take a selfie with the cause btw

1

2

21

Zhengzhong Tu

@_vztu

2 years

Want a *global* and 'fully-conv' MLP model with *linear* complexity that is suitable for hi-res vision? Check out our #CVPR2022 *Oral* paper MAXIM🦾, a general MLP-based backbone, achieving SoTA on denoising, deblurring, deraining, dehazing, low-light enhancing; results👇 1/n🧵

Peyman Milanfar

@docmilanfar

2 years

We've released code and models for MAXIM, our multi-axis MLP backbone for broad use in image processing. w/ @yinxiaoli , @zhengzhongtu , et al. The paper will be an oral presentation at #CVPR2022

1

9

57

1

4

21

Zhengzhong Tu

@_vztu

4 months

Hyperparameter tuning explained

0

2

21

Zhengzhong Tu

@_vztu

5 months

2030: Announcing NVIDIA A1e9 GPU portable tensor core

1

20

Zhengzhong Tu

@_vztu

5 months

This is, wild

0

17

Zhengzhong Tu

@_vztu

2 years

@bboczeng

1

0

18

Zhengzhong Tu

@_vztu

7 months

Help my dear friend @Junfeng27879808 , from University of Maryland to introduce his most recent watermarking work published at @NeurIPSConf 2023. TL;DR: designed a novel and practical watermark framework for dataset copyright protection, that is harmless and effective. Code &…

1

17

Zhengzhong Tu

@_vztu

8 months

@CVL_ETH Please investigate it

0

17

Zhengzhong Tu

@_vztu

6 months

Reviewer 2: "This current figure 2 is too small to read as this is the main methodology figure..., reject"

3

0

16

Zhengzhong Tu

@_vztu

5 months

"AI cannot even dominate humans like cats do." "The same goes for dogs."

1

16

Zhengzhong Tu

@_vztu

2 years

Also @CSProfKGD !

Zhengzhong Tu

@_vztu

2 years

Already back in Austin now! Finally caught up with @ak92501 the Arxiv robot on the last day of CVPR~

0

3

60

0

16

Zhengzhong Tu

@_vztu

7 months

When your paper got rejected twice, for different reasons

1

2

10

Zhengzhong Tu

@_vztu

5 months

State-of-the-art GenAI: Unstable Video Diffusion

0

15

Zhengzhong Tu

@_vztu

2 years

@ylecun @alcinos26 Thanks for sharing our work Dr LeCun🥳! DETR combines CNN & ViT in a "macro" manner, while ours fuses them more "micro-ly": each layer has Conv, local (in Swin), and global (sparse) attention with linear complexity. We've actually put together an AI blog:

A Multi-Axis Approach for Vision Transformer and MLP Models

Posted by Zhengzhong Tu and Yinxiao Li, Software Engineers, Google Research Convolutional neural networks have been the dominant machine learning a...

research.google

0

1

14

Zhengzhong Tu

@_vztu

5 months

A bold start for #NeurIPS 2023 with fish-n-chips and stunning sunset in #NewOrleans !

0

14

Zhengzhong Tu

@_vztu

2 years

Caught the dearest @jaywhang_ ! @2ptmvd @docmilanfar #CVPR2022

mauricio delbracio

@2ptmvd

2 years

Tomorrow, our dearest @jaywhang_ is presenting this very nice work 👇👇 [Oral Session 4.1.2: Computational Photography, early in the morning right after the coffee shot] @CVPR @CVPRConf #CVPR2022

0

3

14

0

13

Zhengzhong Tu

@_vztu

5 months

- What you got for #NeurIPS breakfast? - It’s a hard computer vision problem🤣🤣

2

0

13

Zhengzhong Tu

@_vztu

5 months

So many early-bird registers! #NeurIPS 2023 #NewOrleans

0

13

Zhengzhong Tu

@_vztu

7 months

Subway的烤牛肉加肉全蔬菜全麦三明治可以加入到超值(健康版)穷鬼套餐了。应该是性价比最高的可以吃到新鲜蛋白质和蔬菜的快餐了。��惜在湾区这个物价下，这一份竟然要17$🥵🥵

8

0

10

Zhengzhong Tu

@_vztu

5 months

Our team is also hiring interns & student researchers next year. PhD students working on computer vision and generative AI are welcomed to apply. Feel free to DM or chat with me for this opportunity in NeurIPS! Check out the application link in my previous tweets.

0

1

11

Zhengzhong Tu

@_vztu

2 years

#CVPR2022 Covid update: tested positive this morning, but yesterday was most severe for me with fever and sore throat. It seems if you developed covid-like symptoms, the testing should span several days until the symptoms totally disappear. Anyway, hope everyone can recover soon!

Zhengzhong Tu

@_vztu

2 years

Had a mild sore throat today. Also got informed quite a few friends I have been extensively talking with reported positive. Luckily for me, got both negative results yesterday and today. Extremely fun #CVPR2022 ! For unlucky folks caught Covid, hope y’all a speed recovery!

0

7

2

0

10

Zhengzhong Tu

@_vztu

5 months

Diffusion models Oral Session #NeurIPS2023 . Pretty crazy, huh?

0

10

Zhengzhong Tu

@_vztu

2 years

@bboczeng 重大新闻：曾勃为AutoML指明方向

1

0

10

Zhengzhong Tu

@_vztu

5 months

When you deploy the state-of-the-art model in real world

0

10

Zhengzhong Tu

@_vztu

4 months

Folks, Happy New Year‼️ Let's make the year 2024 the best year so far, and the worst year from now on, for ourselves, family and friends, for AI, and for humanity, CHEERS! ♥️♥️♥️ #NewYear #NewYear2024 #HappyNewYear #HappyNewYear2024 #Cat #NewYearNewMe #Year2024 #dayone