AD Research Hub
DashboardPeoplePapersWorkshopsDatasetsResearch Map
Login
⌘K
AD Research Hub — Anomaly Detection in Computer Vision
← Back to People

Trevor Darrell

University of California, Berkeley

Top CV ResearchersFrontier Research MapScore: 9h-index: 156225,499 citations
HomepageBerkeley homepageUCLA CS 201 talk pageBerkeley research profileSemantic Scholar
Top CV Researcher — Rank #3 (top 10)

Professor; founding co-director of BAIR

Contributions

Caffe, visual recognition, multimodal learning, embodied perception

Why Selected

Long-running leader in visual recognition, multimodal learning, and large-scale vision systems, with strong academic and translational impact.

Score Breakdown

3

historical impact

2

recent visibility

2

current influence

2

asset availability

9

total

Frontier Research Map

Featured Work

The Surprising Efficacy of "Ungrounded" Models for Image and Video Understanding, and Generation

official seminar page — 2024-03-12

Why Now

This is a useful frontier tension: language priors are often shockingly effective even when they are not deeply grounded in vision.

Key Ideas

  • -Language models can contribute nontrivial structure to image and video reasoning even without direct physical grounding.
  • -Visual systems increasingly look like orchestration layers over heterogeneous pretrained modules.
  • -Useful multimodal intelligence may emerge from composition before full grounding is solved.

Open Questions

  • ?Where is the boundary between productive language prior and confident hallucination?
  • ?Should we design multimodal systems as monoliths or as compositional toolchains?
  • ?What tasks punish ungrounded shortcuts strongly enough to force genuine perception?
Canonical CV Leadershigh confidence
Cross-References

Themes

multimodalityLLMsvisual programming

Sign in to access this content

Sign in