Title: Annotation-Efficient Video Representation Learning

Date & Time: Tuesday, October 26th at 12:00 pm

Place: KEC1007 and Zoom Meeting Room.

Abstract: Labeling videos is costly, time-consuming, and tedious. These costs can escalate in applications such as medical diagnosis, etc. where we need domain expertise for annotation. Few-shot learning in Videos aims to solve this problem by building annotation-efficient learning mechanisms. This thesis presents MetaUVFS as the first Unsupervised Meta-learning algorithm for Video Few-Shot action recognition. MetaUVFS leverages over 550K unlabeled videos to train a two-stream 2D and 3D CNN architecture via contrastive learning to capture the appearance-specific spatial and action-specific Spatio-temporal video features respectively. MetaUVFS comprises a novel Action-Appearance Aligned Meta-adaptation~(A3M) module that learns to focus on the action-oriented video features in relation to the appearance features via explicit few-shot episodic meta-learning over unsupervised hard-mined episodes. Our action-appearance alignment and explicit few-shot learner conditions the unsupervised training to mimic the downstream few-shot task, enabling MetaUVFS to significantly outperform all state-of-the-art unsupervised methods on few-shot benchmarks. Moreover, unlike previous few-shot action recognition methods that are supervised, MetaUVFS needs neither base-class labels nor a supervised pre-trained backbone. Thus, we need to train MetaUVFS just once to perform competitively or sometimes even outperform state-of-the-art supervised methods on popular HMDB51, UCF101, and Kinetics100 few-shot datasets.

Committee: Dr. Fuxin Li, Dr. Sinisa Todorovic, Dr. Stefan Lee, Dr. Yelda Turkan