Delta Sampling R-BERT for limited data and low-light action recognition

TL;DR

This paper introduces Delta Sampling R-BERT, a method for action recognition in dark environments that achieves low error rates on limited data by employing novel training strategies and frame selection techniques.

Contribution

The work presents a new approach combining Delta Sampling R-BERT with domain transfer and frame selection for low-light action recognition on small datasets.

Findings

Achieved 11% lower error rate than previous baselines.

Effective in dark environments with limited training data.

Demonstrated on ARID dataset with promising results.

Abstract

We present an approach to perform supervised action recognition in the dark. In this work, we present our results on the ARID dataset. Most previous works only evaluate performance on large, well illuminated datasets like Kinetics and HMDB51. We demonstrate that our work is able to achieve a very low error rate while being trained on a much smaller dataset of dark videos. We also explore a variety of training and inference strategies including domain transfer methodologies and also propose a simple but useful frame selection strategy. Our empirical results demonstrate that we beat previously published baseline models by 11%.