Log inSign up
user avatar
Richard Zhang
@rzhang88
Sr Research Scientist @AdobeResearch PhD @berkeley_ai, BS/MEng @cornellece 🤖 Computer vision, deep learning, graphics
San Francisco, CA
richzhang.github.io
Joined March 2010
302
Following
8,350
Followers
  • user avatar
    Richard Zhang
    @rzhang88
    Jul 19, 2022
    Replying to @waxpancake @minimaxir and @ByFrustrated
    Very neat trick to tease this out. Reproduced: - labs.openai.com/s/7l52BzoeWjpX… - labs.openai.com/s/0bsGqYB35CCm… - labs.openai.com/s/tH77ryI0WO4j… I cherry-picked from ~8 generations, since #dalle #dalle2 is adding a different set of word(s) for each generation
    Image
    Image
    Image
  • user avatar
    Richard Zhang
    @rzhang88
    Sep 9, 2020
    Dear orange picture people: you obviously loaded as BGR by accident. Simply flip the channels and you're good to go 👍 PC: Alexei A. Efros
    Image
    Image
  • user avatar
    Richard Zhang
    @rzhang88
    Nov 4, 2021
    Replying to @AlexTamkin
    This is from buggy implementations in standard libraries, **not a fundamental mathematical issue**. For large factors, buggy implementations are essentially equivalent to naive subsampling. The libraries will be fixed...eventually. Attached image from arxiv.org/abs/2104.11222
    Image
  • user avatar
    Richard Zhang
    @rzhang88
    Sep 6, 2020
    Ask any graphics or signal processing person: *not antialiasing* and ignoring Nyquist sampling theorem is a bug pip install antialiased-cnns import antialiased_cnns model = antialiased_cnns.resnet50(pretrained=True) Now your convnet is antialiased 😃
    Image
    GitHub - adobe/antialiased-cnns: pip install antialiased-cnns to improve stability and accuracy
    From github.com
  • user avatar
    Richard Zhang
    @rzhang88
    Aug 9, 2023
    Amazing GenAI results don't just come from thin air! They're a reflection of the underlying training data. Can we specifically identify the highly influential data? See our ICCV work on Data Attribution w/ @ShengYuWang6, @junyanz89, A. A. Efros: peterwang512.github.io/GenDataAttribu… 1/
    Image
    00:00
    68K
  • user avatar
    Richard Zhang
    @rzhang88
    Jul 29, 2022
    I switched groups in my 3rd year, pushed hard but still missed 2 paper deadlines. The whole PhD thing was not looking very good. After deciding not to submit to CVPR that night, my advisor said "let's chat". I thought I was in trouble...1/2
  • user avatar
    Richard Zhang
    @rzhang88
    Sep 12, 2023
    I was included on MIT @techreview's Innovators Under 35 list (at a youthful 34.9)! technologyreview.com/innovator/rich… My appreciation to the wonderful collaborators, mentors, and labmates at @AdobeResearch and @berkeley_ai for the past decade 1/
    user avatar
    Adobe
    @Adobe
    Sep 12, 2023
    "I'm so grateful to my labmates and mentors, who are world-class researchers," shares Senior Research Scientist @rzhang88 on his name being added to @techreview's prestigious list of top Innovators Under 35 for his achievements in gen AI & image forensics. adobe.ly/45O734E
    Senior Research Scientist Richard Zhang has been named to the Top 35 Innovators Under 35 list by MIT Technology Review
    Image
    Richard Zhang
    From technologyreview.com
    28K
  • user avatar
    Richard Zhang
    @rzhang88
    Oct 27, 2021
    When I started working on this 5 years ago, I often used family photos to test. My grandma (who recently passed) always got a kick out of it. Funnily, this image started propagating into other people's papers/presentations Happy we've improved @Photoshop Colorize! Try it out
    Image
    Image
    Image
    Image
    user avatar
    Jess Magdefrau
    @JessMagdefrau
    Oct 26, 2021
    Over here crying my eyes out with this new color changing Neural Photoshop filter. I colorized my late grandma and grandpa's wedding photo 6 years ago in college. It took me a total of 5hrs to do so. The image below, took 2 minutes with one click of a button #AdobeMAX #photoshop
  • user avatar
    Richard Zhang
    @rzhang88
    Jul 29, 2022
    Replying to @rzhang88
    He said, "Richard, you have all the skills to be a great researcher. Don't worry, keep doing what you're doing, and it will all be okay". A few months later, the colorization paper happened. It all ended up okay. My wonderful advisor and labmates saved my career. 2/2
  • user avatar
    Richard Zhang
    @rzhang88
    Jul 13, 2022
    I'm...IN!!!! #dalle2 "an asian male happy to be granted access to an AI text-to-image program typing in every waking thought he has into it"
    Image
    Image
    Image
  • user avatar
    Richard Zhang
    @rzhang88
    Apr 7, 2021
    Swapping autoencoder code by Taesung Park is out now! NeurIPS '20. Also w/ @junyanz89 @oliverwang81 @JingwanL @elishechtman Alexei Efros Github: github.com/taesungp/swapp… Website: taesung.me/SwappingAutoen… 3 min video: youtube.com/watch?v=0elW11…
    Image
    GitHub - taesungp/swapping-autoencoder-pytorch: Official Implementation of Swapping Autoencoder for...
    From github.com
  • user avatar
    Richard Zhang
    @rzhang88
    Mar 3, 2022
    5 years ago, @junyanz89 taught me how to hotkey Powerpoint alignment keys, and it changed my figure-making life. Happy ECCVing
    Image
  • user avatar
    Richard Zhang
    @rzhang88
    Aug 31, 2020
    Somehow, our ECCV'16 project website got popular and started propagating around richzhang.github.io/colorization/. Four years later, I have gotten around to cleaning it up. Hope folks find it helpful: github.com/richzhang/webp…. Originally made by most helpful postdoc ever, @phillip_isola
    Image
  • user avatar
    Richard Zhang
    @rzhang88
    Jun 11, 2019
    Modern convnets ignore the Nyquist sampling criterion, making them unstable. Come see how simple antialiasing can make your net more stable, accurate, and robust! 3pm tomorrow (Wed) in Seaside Ballroom. bit.ly/2ZjzBAc #ICML2019
    Image
    GIF