"Knights": First Place Submission for VIPriors21 Action Recognition Challenge at ICCV 2021

TL;DR

The paper introduces 'Knights', a novel approach combining self-supervised pretraining, video transformers, and optical flow to achieve top performance in small dataset action recognition at ICCV 2021.

Contribution

It presents a new method that effectively leverages self-supervised learning and multimodal inputs for data-efficient action recognition without external data.

Findings

Achieved 73% accuracy on Kinetics400ViPriors test set.

Outperformed all other entries in the ICCV 2021 challenge.

Demonstrated effectiveness of self-supervised pretraining with video transformers.

Abstract

This technical report presents our approach "Knights" to solve the action recognition task on a small subset of Kinetics-400 i.e. Kinetics400ViPriors without using any extra-data. Our approach has 3 main components: state-of-the-art Temporal Contrastive self-supervised pretraining, video transformer models, and optical flow modality. Along with the use of standard test-time augmentation, our proposed solution achieves 73% on Kinetics400ViPriors test set, which is the best among all of the other entries Visual Inductive Priors for Data-Efficient Computer Vision's Action Recognition Challenge, ICCV 2021.