Selective Feature Compression for Efficient Activity Recognition Inference

TL;DR

This paper introduces Selective Feature Compression (SFC), a method that significantly improves the inference efficiency of action recognition models on trimmed videos by dropping non-informative features without sacrificing accuracy.

Contribution

The paper proposes a novel SFC strategy that compresses spatio-temporal features during inference, reducing computation and memory usage while maintaining or improving accuracy.

Findings

SFC reduces inference time by 6-7x and memory by 5-6x.

SFC slightly improves Top1 accuracy on multiple datasets.

SFC effectively learns to focus on important video regions.

Abstract

Most action recognition solutions rely on dense sampling to precisely cover the informative temporal clip. Extensively searching temporal region is expensive for a real-world application. In this work, we focus on improving the inference efficiency of current action recognition backbones on trimmed videos, and illustrate that one action model can also cover then informative region by dropping non-informative features. We present Selective Feature Compression (SFC), an action recognition inference strategy that greatly increase model inference efficiency without any accuracy compromise. Differently from previous works that compress kernel sizes and decrease the channel dimension, we propose to compress feature flow at spatio-temporal dimension without changing any backbone parameters. Our experiments on Kinetics-400, UCF101 and ActivityNet show that SFC is able to reduce inference speed by 6-7x and memory usage by 5-6x compared with the commonly used 30 crops dense sampling procedure, while also slightly improving Top1 Accuracy. We thoroughly quantitatively and qualitatively evaluate SFC and all its components and show how does SFC learn to attend to important video regions and to drop temporal features that are uninformative for the task of action recognition.