Justin Johnson

Justin Johnson is an American computer scientist and computer-vision researcher, an associate professor at the University of Michigan, and a co-founder of World Labs, the spatial-intelligence company he started in 2024 with his Stanford PhD advisor Fei-Fei Li.
Nextomoro
9 min read
Justin Johnson

Justin Johnson

Justin Johnson is an American computer scientist whose research spans computer vision, machine learning, image generation, and three-dimensional scene understanding. He is an associate professor of computer science and engineering at the University of Michigan (on partial leave) and a co-founder of World Labs, the San Francisco spatial-intelligence company launched in 2024 by his former Stanford PhD advisor Fei-Fei Li. As of May 2026, his commercial role at World Labs is paired with continued doctoral advising at Michigan and the canonical instructional record from Stanford CS231n, the convolutional-neural-networks course whose 2017 lecture videos are among the most-watched computer-vision teaching artifacts on the public web.

At a glance

Origins

Johnson is American and grew up in California. He completed his bachelor's degree at the California Institute of Technology in 2014 with a major in mathematics and computer science, the joint Caltech track that combines a heavy theoretical-mathematics curriculum with the computer-science track in the Computing and Mathematical Sciences division. During the Caltech years he held three consecutive software-engineering internships at Google between 2011 and 2013, and a research internship at Yahoo in summer 2014, foundational industry exposure that preceded the academic computer-vision research career.

Johnson moved to Stanford in 2014 to enter the computer-science PhD program in the Stanford Vision Lab under Fei-Fei Li. The dissertation period coincided with the transition of computer vision from hand-engineered features to deep convolutional networks, and Johnson's collaborator network in the lab included Andrej Karpathy (then a Stanford PhD student under Li before becoming a Tesla and OpenAI engineering leader) and Serena Yeung. He received the Stanford PhD on June 13, 2018, with a dissertation on compositional visual intelligence, the line of research that connects scene graphs, visual question answering, and structured image generation.

Career

The Stanford PhD years from 2014 to 2018 produced Johnson's foundational publication record across visual reasoning, vision-and-language, image generation, and 3D scene understanding, with Fei-Fei Li as advisor and Andrej Karpathy and Serena Yeung as principal lab collaborators. The papers are summarized in Notable contributions; the dissertation, on compositional visual intelligence, was defended in June 2018.

Across 2016 to 2019 Johnson co-taught the Stanford CS231n "Convolutional Neural Networks for Visual Recognition" course with Fei-Fei Li, Andrej Karpathy, and Serena Yeung. The Spring 2017 lecture series, released on YouTube, has stood as one of the most-watched computer-vision teaching artifacts on the public web; the course material was the basis for Johnson's later Michigan curriculum.

After the Stanford PhD Johnson joined Facebook AI Research (FAIR) as a research scientist, beginning in 2018. He held the FAIR role concurrently with the Michigan academic appointment from 2019 onward, a part-time research-scientist arrangement common among FAIR-affiliated computer-vision faculty during the period. The FAIR-era publications include the 2019 Mesh R-CNN paper with Georgia Gkioxari and Jitendra Malik. The FAIR tenure ran approximately to 2023.

Johnson joined the University of Michigan Department of Electrical Engineering and Computer Science as an assistant professor in fall 2019, with the appointment overlapping the FAIR role. The Michigan teaching record includes EECS 498/598 "Deep Learning for Computer Vision" (Fall 2019, Fall 2020, Winter 2022) and EECS 442 "Computer Vision" (Winter 2020 and Winter 2021). The Fall 2019 EECS 498 lectures were posted publicly to YouTube in 2020 and have become a widely cited online curriculum. He was promoted to associate professor in the Michigan EECS department in subsequent years.

In early 2024 Johnson co-founded World Labs with Fei-Fei Li (chief executive officer), Ben Mildenhall (formerly Google, co-author of NeRF), and Christoph Lassner (formerly Meta Reality Labs Research). The founding thesis is that "spatial intelligence," meaning AI systems that understand, generate, and reason about three-dimensional environments, is a research direction separable from the dominant LLM and 2D-image-generation paradigms. Johnson's research record in vision-and-language, scene graphs, and 3D reconstruction matched the thesis. World Labs raised a $230 million seed at a $1 billion-class valuation in September 2024 (led by Andreessen Horowitz, Radical Ventures, and NEA) and a $1 billion strategic round in February 2026 (with Autodesk contributing $200 million as strategic anchor). The first commercial product, Marble, generates and edits persistent 3D environments from text, images, video, or 3D layouts.

Affiliations

Notable contributions

  • Visual Genome (2016, IJCV). Coauthored 108,000-image dataset annotated with object relationships, attributes, and region descriptions; with Ranjay Krishna, Yuke Zhu, Fei-Fei Li, and others. Cumulative citation count places it among the most-cited computer-vision dataset papers of the 2010s.
  • DenseCap: Fully Convolutional Localization Networks for Dense Captioning (CVPR 2016). With Andrej Karpathy and Fei-Fei Li. Introduced the dense-captioning task, in which a model jointly localizes and describes salient regions in an image in natural language.
  • Perceptual Losses for Real-Time Style Transfer and Super-Resolution (ECCV 2016). With Alexandre Alahi and Fei-Fei Li. Established the use of pretrained-network features as a perceptual loss for training feed-forward image-transformation networks; provided the architectural pattern that real-time neural style transfer was built on and that subsequent generative-image work re-used.
  • CLEVR: A Diagnostic Dataset for Compositional Language and Elementary Visual Reasoning (CVPR 2017). With Bharath Hariharan, Laurens van der Maaten, Fei-Fei Li, Larry Zitnick, and Ross Girshick. Provided a synthetic visual-reasoning benchmark that became canonical for evaluating compositional capability in vision-language models.
  • Image Generation from Scene Graphs (CVPR 2018). With Agrim Gupta and Fei-Fei Li. Introduced an end-to-end scene-graph-to-image network using graph convolution and computed scene layouts.
  • Mesh R-CNN (ICCV 2019). With Georgia Gkioxari and Jitendra Malik at FAIR. Extended the Mask R-CNN object-detection architecture with a 3D triangle-mesh head for joint detection and 3D shape prediction; a foundational reference for the 3D-vision research line that informs the World Labs program.
  • Stanford CS231n course co-instruction with Fei-Fei Li, Andrej Karpathy, and Serena Yeung (2016 to 2019). The Spring 2017 lecture series, released on YouTube, is among the most-watched computer-vision teaching artifacts publicly available.
  • Michigan EECS 498/598 "Deep Learning for Computer Vision." Johnson's evolution of the CS231n curriculum at Michigan, with the Fall 2019 edition released in full on YouTube and updated topical coverage including transformers, video, and 3D reasoning.
  • World Labs co-founding and the spatial-intelligence research program (2024 onward). Operationalization of the 3D-scene-understanding research line as a commercial program in San Francisco, with Marble as the first commercial product.

Investments and boards

The entries below are limited to AI, semiconductors, datacenters, software, and energy.

  • World Labs (AI): Co-founder, 2024 to present. $230 million seed at a $1 billion-class valuation, September 2024 (Andreessen Horowitz, Radical Ventures, NEA leads); $1 billion strategic round in February 2026 with Autodesk contributing $200 million as strategic anchor.

No personal angel investments in AI, semiconductors, datacenters, software, or energy companies have been publicly disclosed as of May 2026.

Network

Johnson's longest-running professional relationship is with Fei-Fei Li, his Stanford PhD advisor and now World Labs chief executive officer; the advisor-and-student relationship has run continuously from 2014 through the World Labs co-founding in 2024 and the early product launches. Andrej Karpathy, an earlier Li PhD student and CS231n co-instructor (2011 to 2015), overlapped the Stanford Vision Lab years and remained a peer in the broader computer-vision-and-large-models community through the OpenAI and Tesla periods. Serena Yeung, the third 2017 CS231n co-instructor, is a Stanford computer-science assistant professor and frequent collaborator on the visual-reasoning research line.

The World Labs co-founder team produces Johnson's primary current commercial collaborators: Ben Mildenhall, formerly a Google research scientist and co-author of the NeRF paper that catalyzed the modern 3D-reconstruction research direction; and Christoph Lassner, formerly senior researcher at Meta Reality Labs Research and contributor of graphics and 3D-reconstruction expertise. The four-co-founder team is the operating core of the spatial-intelligence research program.

The FAIR research-scientist tenure produced a separate collaborator network anchored by Georgia Gkioxari (now at Caltech) and Jitendra Malik (at Berkeley and FAIR), the Mesh R-CNN co-authors; and a broader FAIR-vision peer group during the Mesh R-CNN and PyTorch3D research lines. The Michigan EECS computer-vision faculty, including Jia Deng (later moved to Princeton), provides additional academic-research network depth.

Position in the field

Johnson is unusual in the senior computer-vision cohort for combining three distinct credentials at once: a research record across the foundational dataset, vision-and-language, and 3D-reconstruction lines published from 2014 onward; an instructional reach through Stanford CS231n and Michigan EECS 498 that has trained much of the broader applied-deep-learning practitioner community; and a current operating role as a co-founder of a venture-funded commercial AI lab, World Labs, with the spatial-intelligence thesis.

His Google Scholar citation count exceeds 130,000 across the published research record, with concentrations in Visual Genome, DenseCap, perceptual losses, CLEVR, Image Generation from Scene Graphs, and Mesh R-CNN. The Stanford CS231n and Michigan EECS 498 lecture videos are among the most-cited public-curriculum resources in computer vision; industry coverage of the canonical applied-deep-learning lecture series consistently names Johnson, alongside Fei-Fei Li and Andrej Karpathy, as one of the principal instructional voices of the modern computer-vision era.

The World Labs role frames Johnson's current public posture as a working co-founder of a commercial frontier-research lab whose thesis aligns with his published research record. The senior computer-vision academic researchers who have made comparable transitions to commercial AI labs include Yann LeCun (NYU and Meta to AMI Labs), Jitendra Malik (Berkeley and FAIR), and Geoffrey Hinton (Toronto and Google); Johnson's transition is at an earlier career stage but follows the same pattern of academic-research credentials informing a focused commercial research program.

Outlook

Open questions over the next 6 to 18 months:

  • World Labs research and product cadence. Whether Johnson's research output through the World Labs research line continues at the publication cadence of the Stanford and FAIR years, and whether World Labs follows Marble with successor 3D-scene-generation products and open research artifacts.
  • Stanford CS231n successor curriculum. Continued maintenance and online release of the Michigan EECS 498 "Deep Learning for Computer Vision" course as the canonical successor to the original Stanford CS231n.
  • Michigan academic continuity. Whether Johnson remains at Michigan on partial leave through the World Labs commercial period or transitions to full-time at World Labs as the spatial-intelligence commercial program scales.
  • Spatial-intelligence research direction. Whether the World Labs research thesis (3D scene representation as a separable capability domain from LLMs) is supported by published benchmarks and follow-on research artifacts; and whether Johnson's publications continue to anchor the program with new architectural results.
  • PhD-student pipeline. Continuity of doctoral advising at Michigan during the World Labs period, and whether Johnson's lab continues to graduate computer-vision PhDs into either World Labs or the broader AI-lab ecosystem.

Sources

Person World Labs
About the author
Nextomoro

Nextomoro

nextomoro tracks progress for AI research labs, models, and what's next.

Read next

Ekin Dogus Cubuk

Andrew White

Daniel Balsam

Laurent Massoulié

Patrick Pérez

AI Research Lab Intelligence

nextomoro tracks progress for AI research labs, models, and what's next.

AI Research Lab Intelligence

Great! You’ve successfully signed up.

Welcome back! You've successfully signed in.

You've successfully subscribed to AI Research Lab Intelligence.

Your link has expired.

Success! Check your email for magic link to sign-in.

Success! Your billing info has been updated.

Your billing was not updated.