L3D-IVU CVPR2023
2nd Workshop on Learning with Limited Labelled Data for Image and Video Understanding
19 June 2023
Location: Vancouver Convention Center
East 3
motivation
Deep learning has been widely successful in a variety of computer vision tasks such as object recognition, object detection, and semantic segmentation. It also has been deployed with success in learning spatiotemporal features for video segmentation/detection and action recognition tasks. However, one of the major bottlenecks of deep learning in both image and video understanding tasks is the need for large-scale labelled datasets. Collecting and annotating such datasets can be labor intensive and costly. In many scenarios of practical interest only a few labelled examples of novel categories may be available at model training time. Currently available large-scale data typically cover relatively narrow sets of categories and are constrained by licensing. As such, they are often hard to naively apply to practical problems. It is especially problematic in developing countries that do not have the required resources to collect large scale labelled datasets for new tasks. The goal of this workshop is to explore approaches that learn from limited labelled data, or with side information such as text data, or using data with weak/self supervision, with special focus on video understanding tasks. This will be the second L3D-IVU workshop in conjunction with CVPR, where it had a great success and wide interest from multiple researchers as it explores the intersection of learning with limited labelled data and video understanding.
News and Updates:
Jan 10, 2023: Workshop Officially Confirmed for CVPR 2023, that will take place on Monday June 19, 2023.
Jan 13, 2023: Paper Submission is open at, https://cmt3.research.microsoft.com/L3DIVU2023.
Feb 17, 2023: Paper submission deadline has been extended till March 6.
Feb 21, 2023: Challenge details on Long tail visual relationship recognition released.
April 3, 2023: Author notifications are sent.
April 15, 2023: A change in our speakers schedule, Gabriela Csurka and Eleni Triantafillou apologized for not being able to attend in-person.
April 22, 2023: We are very excited to have Associate Prof. Ismail Ben Ayed and Naila Murray to be speakers in our workshop.
May 7, 2023: Our LTVRR Challenge deadline is postponed till June 1st, 2023.
June 12, 2023: A social and mixer will happen after the workshop as a BBQ and board games event.
Challenge Prizes:
1st Winner: IPad (Apple)
Best Paper prize:
1000$ in value (Borealis AI)
Best Paper Prize, awarded by Eirene Seiradaki, Director of Research Partnerships at Borealis AI.
(Only students/researchers are eligible to participate. Government officials, public sector officials, and employees of entities who do business in the Public Sector are not eligible to participate. Apple and Borealis AI reserves the right not to make the award if the recipient turns out to be a government official, public sector official, or individual who does business in the public sector.)
Participation:
We encourage submissions that are under one of the topics of interest, but also we welcome other interesting and relevant research for learning with limited labeled data.
Few-Shot classification, detection and segmentation in still images and video, including objects, actions, scenes and object tracking.
Zero-shot learning in video understanding.
Video and language modelling.
Self supervised Learning in video related tasks.
Weakly/Semi supervised learning in video understanding.
Transfer Learning.
Open-set learning.
New benchmarks and metrics.
Real-world applications discussing the societal impact of few-shot learning.
Papers will be peer-reviewed under a double-blind policy and the submission deadline is the 8th of February 2023. Accepted papers will be presented at the poster session, some as orals, and one paper will be awarded as the best paper.
Invited Speakers
Naila Murray
Meta AI
Hazel Doughty
University of Amesterdam
Ismail Ben Ayed
ETS MontrealChristoph Feichtenhofer
Meta AI
Trevor Darrell
UC Berkley
Graham Taylor
University of Guelph, Borealis AI