Low-Resolution Action Recognition for Tiny Actions Challenge

TL;DR

This paper presents a comprehensive approach for recognizing tiny actions in surveillance videos, addressing challenges of low resolution and long-tailed data distribution through data balancing, dual-resolution distillation, and model ensembling, achieving top performance.

Contribution

It introduces a dual-resolution distillation framework and a data balancing strategy specifically designed for tiny action recognition in surveillance videos.

Findings

Achieved Top-1 ranking on the challenge leaderboard.

Effectively alleviated data bias with data balancing.

Enhanced low-resolution recognition via super-resolution knowledge distillation.

Abstract

Tiny Actions Challenge focuses on understanding human activities in real-world surveillance. Basically, there are two main difficulties for activity recognition in this scenario. First, human activities are often recorded at a distance, and appear in a small resolution without much discriminative clue. Second, these activities are naturally distributed in a long-tailed way. It is hard to alleviate data bias for such heavy category imbalance. To tackle these problems, we propose a comprehensive recognition solution in this paper. First, we train video backbones with data balance, in order to alleviate overfitting in the challenge benchmark. Second, we design a dual-resolution distillation framework, which can effectively guide low-resolution action recognition by super-resolution knowledge. Finally, we apply model en-semble with post-processing, which can further boost per-formance on the long-tailed categories. Our solution ranks Top-1 on the leaderboard.