Richard Zhang

Career

Zhang worked as a staff engineer at the Johns Hopkins University Applied Physics Laboratory from 2010 to 2012 and then completed a PhD in EECS at UC Berkeley in 2018, advised by Alexei A. Efros, with a thesis on image synthesis for self-supervised visual representation learning. Zhang interned at Adobe Research in 2017 and joined it as a research scientist in May 2018, becoming senior research scientist in 2022 and later principal scientist.⁷⁸¹

Research

Zhang's earlier work spans self-supervised learning, perceptual metrics and image forensics. It includes Colorful Image Colorization (ECCV 2016), the LPIPS perceptual similarity metric (CVPR 2018) and "Making Convolutional Networks Shift-Invariant Again" (ICML 2019), along with a CVPR 2022 study of FID calculation linked to the clean-fid package for evaluating generative models.⁷⁵²⁹¹

Zhang's recent research focuses on controllability, inference speed and data transparency for generative models. Examples include GigaGAN (CVPR 2023), one-step diffusion with distribution matching distillation (CVPR 2024) and its improved version (NeurIPS 2024), data attribution for text-to-image models, and fast causal video generation (CVPR 2025).⁷¹⁰¹

Work and contributions

• LPIPS (PerceptualSimilarity) Open-source code

  First author, with Phillip Isola, Alexei A. Efros, Eli Shechtman, and Oliver Wang, of the CVPR 2018 paper introducing the learned perceptual metric. The repository, released under a BSD-style license that names Zhang among the copyright holders, provides the metric through the lpips Python package.²³⁴¹

• Colorful Image Colorization Open-source code

  First author, with Phillip Isola and Alexei A. Efros, of the ECCV 2016 paper on automatic image colorization; the accompanying repository is released under a BSD-style license that names the three authors as copyright holders.⁵⁶¹

• Antialiased CNNs Published research

  Sole author of "Making Convolutional Networks Shift-Invariant Again" (ICML 2019), which adds anti-aliasing to downsampling layers; the method is distributed as the antialiased-cnns package.⁹¹

Timeline

1. Jul 2010Began work as a staff engineer at the Johns Hopkins University Applied Physics Laboratory.⁷
2. Mar 28, 2016Released "Colorful Image Colorization" as first author (ECCV 2016).⁵
3. Jan 11, 2018Released the LPIPS paper "The Unreasonable Effectiveness of Deep Features as a Perceptual Metric".²
4. May 2018Completed a PhD at UC Berkeley and joined Adobe Research as a research scientist.⁷
5. Apr 25, 2019Released "Making Convolutional Networks Shift-Invariant Again" as sole author (ICML 2019).⁹
6. Jan 2022Became a senior research scientist at Adobe Research.⁷
7. 2023Named to MIT Technology Review's list of Innovators Under 35.¹⁸
8. Nov 30, 2023Co-authored "One-step Diffusion with Distribution Matching Distillation" (CVPR 2024).¹⁰

Other documented roles

Roles the sources document that are not confirmed as current.

• Staff engineer (2010–2012), Johns Hopkins University Applied Physics Laboratory⁷

Honors and recognition

• Innovators Under 35 · MIT Technology Review, 2023¹⁸
• Adobe Research Fellowship · Adobe Research, 2017⁸

Where they work

• Adobe 4 people in the wiki

Related people

• Jun-Yan ZhuAlso at Adobe
• Josef SivicAlso at Adobe
• Christian CantrellAlso at Adobe
• Sanja FidlerComputer vision · Image, video, and 3D generation
• Tero KarrasImage, video, and 3D generation · Computer vision
• Michael J. BlackComputer vision · Image, video, and 3D generation
• Kaiming HeComputer vision · Image, video, and 3D generation
• Alexei A. EfrosComputer vision · Image, video, and 3D generation
• Sayak PaulImage, video, and 3D generation · Computer vision
• Ming-Yu LiuImage, video, and 3D generation · Computer vision