Scalable Frame Sampling for Video Classification: A Semi-Optimal Policy Approach with Reduced Search Space

TL;DR

This paper proposes a semi-optimal frame sampling policy for video classification that reduces the search space from combinatorial to linear, enabling efficient and stable selection of key frames across diverse datasets.

Contribution

It introduces a novel semi-optimal policy that simplifies frame sampling by independently estimating frame importance, significantly reducing computational complexity.

Findings

Reduces search space from O(T^N) to O(T)

Achieves stable high performance across various datasets and models

Efficiently approximates the optimal sampling policy

Abstract

Given a video with $T$ frames, frame sampling is a task to select $N \ll T$ frames, so as to maximize the performance of a fixed video classifier. Not just brute-force search, but most existing methods suffer from its vast search space of $\binom{T}{N}$, especially when $N$ gets large. To address this challenge, we introduce a novel perspective of reducing the search space from $O(T^N)$ to $O(T)$. Instead of exploring the entire $O(T^N)$ space, our proposed semi-optimal policy selects the top $N$ frames based on the independently estimated value of each frame using per-frame confidence, significantly reducing the computational complexity. We verify that our semi-optimal policy can efficiently approximate the optimal policy, particularly under practical settings. Additionally, through extensive experiments on various datasets and model architectures, we demonstrate that learning our semi-optimal policy ensures stable and high performance regardless of the size of $N$ and $T$.