t-EVA: Time-Efficient t-SNE Video Annotation

TL;DR

t-EVA is a novel video annotation method that leverages spatio-temporal features and t-SNE to significantly accelerate the annotation process while maintaining high classification accuracy.

Contribution

The paper introduces a time-efficient annotation approach combining feature similarity and t-SNE for improved video labeling speed.

Findings

Outperforms existing annotation tools in speed

Maintains high classification accuracy

Effective on large-scale video datasets

Abstract

Video understanding has received more attention in the past few years due to the availability of several large-scale video datasets. However, annotating large-scale video datasets are cost-intensive. In this work, we propose a time-efficient video annotation method using spatio-temporal feature similarity and t-SNE dimensionality reduction to speed up the annotation process massively. Placing the same actions from different videos near each other in the two-dimensional space based on feature similarity helps the annotator to group-label video clips. We evaluate our method on two subsets of the ActivityNet (v1.3) and a subset of the Sports-1M dataset. We show that t-EVA can outperform other video annotation tools while maintaining test accuracy on video classification.