VAND: Visual Anomaly and Novelty Detection
CVPR 2023 Workshop
June 18, 2023 in Vancouver, Canada (In Person) + Zoom (Virtual)
Overview
Anomaly detection, and the synonymous topics of novelty and out-of-distribution detection, represent an important and application-relevant challenge within both computer vision and the broader field of pattern recognition. In its simplest formulation, anomaly detection targets the identification of samples which deviate from an obtained approximation to the true distribution of normality for a given dataset. As such anomalies represent unexpected eventualities or outliers in the scope of a given task. The notion of detecting them effectively and efficiently has been sought after for many real-world applications including medical diagnosis, airport security screening, industrial inspection, or crowd control.
However, anomaly detection is far from a simple task due to the challenges of accounting for all forms with which an anomaly may be present. It is typically impossible for any given dataset to account for the complete anomalous variability as they represent an unbounded (open set) distribution of possible deviations from the distribution of normality. Established supervised techniques are therefore prone to suffer from heavy classification bias or over-fitting.
To these ends, we now see the rise of a complex and vibrant set of learning-based paradigms addressing the anomaly detection task - varying across both the fully/semi/un-supervised and few/one/zero shot axes of recent computer vision and pattern recognition research. This workshop brings together researchers of both industry and academia to present and discuss recent developments, opportunities and open challenges in this area. The workshop will also host a challenge for zero-/few-normal-shot anomaly detection, to encourage the development and benchmarking new algorithms for realistic yet challenging tasks.
Invited Speakers
Call for Papers
Our call for papers includes the following topics:
Anomaly detection, novelty detection, and out-of-distribution detection in images and videos.
Relevant learning paradigms including unsupervised, few-shot and active learning.
Dataset challenges including highly imbalanced data, noisy/incomplete labels, data sampling, applications spanning vision-based industrial inspection, predictive maintenance of complex machines.
Adjacent domains such as in-field inspection and medical diagnosis.
Theoretical contributions that address challenges unique to anomaly detection and novelty detection.
Submissions to the workshop must be of 8 page papers, with unlimited space for references and supplementary materials, following the CVPR 2023 style and formatting guidelines. The review process is double-blind and there is no rebuttal. Submissions must not have been previously published in a substantially similar form. Accepted papers will be invited for either spotlight talks or poster presentations. Accepted papers will be published in conjunction with CVPR 2023 proceedings.
Important Dates:
Submission deadline: March 6th, 2023 March 8th, 2023
Author notification: April 3rd, 2023
Camera-ready deadline: April 8th, 2023
Workshop: June 18th, 2023
Please check the Call for Papers page for more details.
Challenges
Welcome to the Visual Anomaly and Novelty Detection (VAND) 2023 Challenge! This year our challenge aims to bring visual anomaly detection closer to industrial visual inspection, which has wide real-world applications. We look forward to participation from both academia and industry.
For industrial visual inspection, the majority of previous methods focus on training a specific model for each category given a large amount of normal images as reference. However, in real-world scenarios, there are millions of industrial products and it is not cost-effective to collect a large training set for each object and deploy different models for different categories. In fact, building cold-start models, models trained with zero or few normal images, is essential in many cases as defects are rare with a wide range of variations.
Building a single model that can be rapidly adapted to numerous categories without or with only a handful of normal reference images is an ideal solution and an open challenge to the community. To encourage the research in this direction, we propose two relevant tracks:
Track 1: Zero-shot Anomaly Detection (Anomaly classification + Segmentation)
Track 2: Few-Shot Anomaly Detection (Anomaly classification + Segmentation)
Note that in both tracks, there will be no training examples of defective examples. We will have two phases and each phase has different test datasets. The first phase aims to kick-start research and development for the given tasks with public datasets. The second phase will release a new test set and we will announce winners according to the results in phase 2.
Please check the Challenge page for more details
Organizing Team
University of Freiburg
Durham University
MedUni Wien
MVTec
AWS AI Labs
Challenge Track
AWS AI Labs
AWS AI Labs
AWS AI Labs
AWS AI Labs
Program Committee
Amir Atapour-Abarghouei (Durham University)
Aditya Deshpande (AWS AI Labs)
Alex Mackin (Amazon Prime Video)
Bodo Rosenhahn (University of Hannover)
Chun-Liang Li (Google)
David Zimmerer (DKFZ)
Dong Gong (UNSW Sydney)
Dongqing Zhang (AWS AI Labs)
Giacomo Boracchi (Polimi)
Guansong Pang (SMU)
Jack Barker (Samsung AI)
Jihoon Tack (KAIST)
Jongheon Jeong (KAIST)
Karsten Roth (University of Tuebingen)
Lukas Ruff (Aignostics)
Marius Kloft (TU Kaiserslautern)
Neelanjan Bhowmik (Durham University)
Oliver Rippel (Scortex)
Peter Gehler (AWS AI Labs)
Raghav Chalapathy (Walmart Labs)
Samet Akcay (Intel)
Thomas Dietterich (Oregon State University)
Yona Falinie Abd. Gaus (Durham University)
Zhisheng Xiao (AWS AI Labs)
Contact: vand-cvpr2023@googlegroups.com