Egocentric Video-Language Pretraining @ EPIC-KITCHENS-100 Multi-Instance Retrieval Challenge 2022

TL;DR

This paper introduces an egocentric video-language pretraining approach leveraging the Ego4D dataset, achieving strong multi-instance retrieval performance on EPIC-KITCHENS-100 with novel training and inference techniques.

Contribution

It pioneers egocentric VLP from pretraining dataset, objective, and development set, and develops a model with adaptive loss and dual-softmax for MIR tasks.

Findings

Achieved 47.39% mAP on challenge test set

Developed an adaptive multi-instance max-margin loss

Utilized dual-softmax for reliable inference

Abstract

In this report, we propose a video-language pretraining (VLP) based solution \cite{kevin2022egovlp} for the EPIC-KITCHENS-100 Multi-Instance Retrieval (MIR) challenge. Especially, we exploit the recently released Ego4D dataset \cite{grauman2021ego4d} to pioneer Egocentric VLP from pretraining dataset, pretraining objective, and development set. Based on the above three designs, we develop a pretrained video-language model that is able to transfer its egocentric video-text representation to MIR benchmark. Furthermore, we devise an adaptive multi-instance max-margin loss to effectively fine-tune the model and equip the dual-softmax technique for reliable inference. Our best single model obtains strong performance on the challenge test set with 47.39% mAP and 61.44% nDCG. The code is available at https://github.com/showlab/EgoVLP.