Featureless: Bypassing feature extraction in action categorization

TL;DR

This paper presents a novel action categorization method that bypasses traditional feature extraction by directly learning to predict video representations from raw data, leveraging a discriminative Waldboost model for efficient classification.

Contribution

It introduces a featureless approach for action recognition that directly predicts video representations from raw data, improving efficiency and flexibility over traditional feature-based methods.

Findings

Achieves competitive accuracy on UCF11 dataset

Demonstrates computational efficiency comparable to feature-based methods

Supports both 2D and 3D video representation prediction

Abstract

This method introduces an efficient manner of learning action categories without the need of feature estimation. The approach starts from low-level values, in a similar style to the successful CNN methods. However, rather than extracting general image features, we learn to predict specific video representations from raw video data. The benefit of such an approach is that at the same computational expense it can predict 2 D video representations as well as 3 D ones, based on motion. The proposed model relies on discriminative Waldboost, which we enhance to a multiclass formulation for the purpose of learning video representations. The suitability of the proposed approach as well as its time efficiency are tested on the UCF11 action recognition dataset.