# Towards Universal Representation for Unseen Action Recognition

## TL;DR

This paper introduces a universal visual-semantic representation for unseen action recognition that generalizes across datasets, leveraging large-scale training data and a novel GMIL framework, leading to significant performance improvements.

## Contribution

It proposes a universal representation framework for UAR that generalizes across datasets using GMIL and semantic adaptation, without additional training.

## Key findings

- Significant improvements on UCF101 and HMDB51 benchmarks.
- Effective generalization to cross-dataset scenarios.
- Utilizes large-scale ActivityNet data for building blocks.

## Abstract

Unseen Action Recognition (UAR) aims to recognise novel action categories without training examples. While previous methods focus on inner-dataset seen/unseen splits, this paper proposes a pipeline using a large-scale training source to achieve a Universal Representation (UR) that can generalise to a more realistic Cross-Dataset UAR (CD-UAR) scenario. We first address UAR as a Generalised Multiple-Instance Learning (GMIL) problem and discover 'building-blocks' from the large-scale ActivityNet dataset using distribution kernels. Essential visual and semantic components are preserved in a shared space to achieve the UR that can efficiently generalise to new datasets. Predicted UR exemplars can be improved by a simple semantic adaptation, and then an unseen action can be directly recognised using UR during the test. Without further training, extensive experiments manifest significant improvements over the UCF101 and HMDB51 benchmarks.