Inductive and Transductive Few-Shot Video Classification via Appearance and Temporal Alignments

TL;DR

This paper introduces a novel few-shot video classification method that combines appearance and temporal alignments, including the first transductive approach, demonstrating improved performance on key datasets.

Contribution

It presents a new framework integrating appearance and temporal alignments for few-shot video classification, including the first transductive method, with extensive experimental validation.

Findings

Appearance and temporal alignments are crucial for temporal order-sensitive datasets.

The proposed method achieves comparable or better results than previous approaches.

The approach is effective on Kinetics and Something-Something V2 datasets.

Abstract

We present a novel method for few-shot video classification, which performs appearance and temporal alignments. In particular, given a pair of query and support videos, we conduct appearance alignment via frame-level feature matching to achieve the appearance similarity score between the videos, while utilizing temporal order-preserving priors for obtaining the temporal similarity score between the videos. Moreover, we introduce a few-shot video classification framework that leverages the above appearance and temporal similarity scores across multiple steps, namely prototype-based training and testing as well as inductive and transductive prototype refinement. To the best of our knowledge, our work is the first to explore transductive few-shot video classification. Extensive experiments on both Kinetics and Something-Something V2 datasets show that both appearance and temporal alignments are crucial for datasets with temporal order sensitivity such as Something-Something V2. Our approach achieves similar or better results than previous methods on both datasets. Our code is available at https://github.com/VinAIResearch/fsvc-ata.