Timezone: »

 
Building One-class Detector for Anything: Open-vocabulary Zero-shot OOD Detection Using Text-image Models
Yunhao Ge · Jie Ren · Jiaping Zhao · Kaifeng Chen · Andrew Gallagher · Laurent Itti · Balaji Lakshminarayanan

@
in Knowledge and Logical Reasoning in the Era of Data-driven Learning »
Event URL: https://github.com/gyhandy/One-Class-Anything »

We focus on the challenge of out-of-distribution (OOD) detection in deep learning models, a crucial aspect in ensuring reliability. Despite considerable effort, the problem remains significantly challenging in deep learning models due to their propensity to output over-confident predictions for OOD inputs. We propose a novel one-class open-set OOD detector that leverages text-image pre-trained models in a zero-shot fashion and incorporates various descriptions of in-domain and OOD. Our approach is designed to detect anything not in-domain and offers the flexibility to detect a wide variety of OOD, defined via fine- or coarse-grained labels, or even in natural language. We evaluate our approach on challenging benchmarks including large-scale datasets containing fine-grained, semantically similar classes, distributionally shifted images, and multi-object images containing a mixture of in-domain and OOD objects. Our method shows superior performance over previous methods on all benchmarks.

[ Poster] 

Author Information

Yunhao Ge (University of Southern California)

I am a Ph.D. Candidate in the CS Department at University of Southern California, advised by Prof. Laurent Itti. I am also a Visiting PhD Student at Stanford Vision and Learning Lab (SVL) advised by Prof. Jiajun Wu. I'm interested in how could human efficiently teach AI to learn the human ability to perceive, understand, interact, and reason the physical world. My current research focuses include: Human-inspired Learning Algorithm (Vision‑Language Models, Lifelong Learning, Visual Reasoning) Learning from Synthetic Data (Sim2Real): using neural renderer (NeRF, Stable Diffusion) and simulation to synthesize realistic and physical plausible data to solve real-world Computer Vision and Robotics problems with minimal human supervision Reliable Deep Learning (Robustness, Out-of-distribution (OOD) Detection, Interpretability)

Jie Ren (Google Brain)
Jiaping Zhao (Google Inc.)
Kaifeng Chen (Google)
Andrew Gallagher (Google)
Laurent Itti (University of Southern California)
Balaji Lakshminarayanan (Google Brain)

More from the Same Authors