Multi-Modal Few-Shot Temporal Action Detection

TL;DR

This paper introduces a novel multi-modal few-shot temporal action detection method called MUPPET, which leverages support videos and class names using vision-language models to improve detection performance on benchmark datasets.

Contribution

The paper proposes MUPPET, a new approach that combines few-shot and zero-shot learning for temporal action detection using multi-modal prompts and meta-learning techniques.

Findings

MUPPET outperforms state-of-the-art methods on ActivityNetv1.3 and THUMOS14 datasets.

MUPPET achieves state-of-the-art results in few-shot object detection on MS-COCO.

The method effectively handles intra-class variation through query feature regulation.

Abstract

Few-shot (FS) and zero-shot (ZS) learning are two different approaches for scaling temporal action detection (TAD) to new classes. The former adapts a pretrained vision model to a new task represented by as few as a single video per class, whilst the latter requires no training examples by exploiting a semantic description of the new class. In this work, we introduce a new multi-modality few-shot (MMFS) TAD problem, which can be considered as a marriage of FS-TAD and ZS-TAD by leveraging few-shot support videos and new class names jointly. To tackle this problem, we further introduce a novel MUlti-modality PromPt mETa-learning (MUPPET) method. This is enabled by efficiently bridging pretrained vision and language models whilst maximally reusing already learned capacity. Concretely, we construct multi-modal prompts by mapping support videos into the textual token space of a vision-language model using a meta-learned adapter-equipped visual semantics tokenizer. To tackle large intra-class variation, we further design a query feature regulation scheme. Extensive experiments on ActivityNetv1.3 and THUMOS14 demonstrate that our MUPPET outperforms state-of-the-art alternative methods, often by a large margin. We also show that our MUPPET can be easily extended to tackle the few-shot object detection problem and again achieves the state-of-the-art performance on MS-COCO dataset. The code will be available in https://github.com/sauradip/MUPPET