Transferring and Adapting Source Knowledge in Computer Vision and VisDA Challenge

In conjunction with ECCV 2020

TASKCV-2020

This is the 7th annual workshop that brings together computer vision researchers interested in domain adaptation and knowledge transfer techniques. Given the last year success, we will keep the Domain Adaptation Challenge.

A key ingredient of the recent successes of computer vision methods is the availability of large sets of annotated data. However, collecting them is prohibitive in many real applications and it is natural to search for alternative source of knowledge that needs to be transferred or adapted to provide sufficient learning support. Our workshop aims to bring together researchers in various sub-areas of Transfer Learning (TL) and Domain Adaptation (DA) for computer vision. Moreover, for this year we propose a new edition of the VisDA challenge that will focus on domain adaptive instance retrieval, where the source and target samples will be drawn from synthetic and real image domains of pedestrians and vehicles.

Session 1: Sunday 23 August - 10:00 - 12:00 UTC+1 Session 2: Sunday 23 August - 20:00 - 22:00 UTC+1


Topics

Accordingly, TASK-CV aims to bring together research in transfer learning and domain adaptation for computer vision and invites the submission of research contributions on the following topics:

  • TL/DA learning methods for challenging paradigms like unsupervised, incremental, open set, universal, online and federated learning.

  • TL/DA CNN architectures with new adaptation techniques, fine-tuning strategies, regularization approaches, weights transfer solutions etc.

  • TL/DA focusing on specific computer vision tasks (e.g., image classification, object detection, semantic segmentation, retrieval, tracking, etc.) and applications (biomedical, robotics, multimedia, autonomous driving, etc.).

  • TL/DA methods working at feature and pixel (generative) level as well as jointly applied with other learning paradigms such as reinforcement learning.

  • DA in case of sensor differences (e.g., low-vs-high resolution, power spectrum sensitivity, different RGB/Depth modalities) and compression schemes.

  • Datasets and protocols for evaluating TL/DA methods.

  • Going beyond TL/DA towards domain generalization.

  • Multi-Task, Zero- One- and Few-Shot Learning.

This is not a closed list; thus, we welcome other interesting and relevant research for TASK-CV.

DOMAIN ADAPTATION CHALLENGE

Please see the challenge website for details, dates, and submission guidelines.

ACCEPTED PAPERS

  • Class-imbalanced Domain Adaptation: An Empirical Odyssey. Shuhan Tan, Xingchao Peng, Kate Saenko

  • Sequential Learning for Domain Generalization. Da Li, Yongxin Yang, Yi-Zhe Song, Timothy Hospedales

  • Generating Visual and Semantic Explanations with Multi-task Network. Wenjia Xu, Jiuniu Wang, Yang Wang, Yirong Wu, Zeynep Akata

  • SpotPatch: Parameter-Efficient Transfer Learning for Mobile Object Detection. Keren Ye, Adriana Kovashka, Mark Sandler, Menglong Zhu, Andrew Howard, Marco Fornoni

  • Using Sentences as Semantic Representations in Large Scale Zero-Shot Learning. Yannick Le Cacheux, Herve Le Borgne, Michel Crucianu

  • Adversarial Transfer of Pose Estimation Regression. Boris Chidlovskii, Assem Sadek

  • Disentangled Image Generation for Unsupervised Domain Adaptation. Safa Cicek, Ning Xu, Zhaowen Wang, Hailin Jin, Stefano Soatto

  • Domain Generalization using Shape Representation. Narges Honarvar Nazari, Adriana Kovashka

  • Bi-Dimensional Feature Alignment for Cross-Domain Object Detection. Zhen Zhao, Yuhong Guo, Jieping Ye

  • Bayesian Zero-Shot Learning. Sarkhan Badirli, Zeynep Akata, Murat Dundar

  • Self-Supervision for 3D Real-World Challenges. Antonio Alliegro, Davide Boscaini, Tatiana Tommasi

  • Diversified Mutual Learning for Deep Metric Learning. Wonpyo Park, Wonjae Kim, Kihyun You, Minsu Cho

  • Domain Generalization vs Data Augmentation: an unbiased perspective. Francesco Cappio Borlino, Antonio D'Innocente, Tatiana Tommasi

ACCEPTED 'ONGOING' WORKS

  • Active Domain Adaptation via Clustering Uncertainty-weighted Embeddings. Viraj Prabhu, Arjun Chandrasekaran, Kate Saenko, Judy Hoffman

  • Multi-Task Incremental Learning for Object Detection. Xialei Liu, Hao Yang, Avinash Ravichandran, Rahul Bhotika, Stefano Soatto

  • Bookworm continual learning: beyond zero-shot learning and continual learning. Kai Wang, Luis Herranz, Anjan Dutta, Joost van de Weijer


The Best Paper Award will be announced during the last Live Workhsop Session.

Sponsors