Contrastive Learning of Global-Local Video Representations

This repository contains PyTorch implementation of the Global-Local network proposed in our paper [Contrastive Learning of Global-Local Video Representations (NeurIPS 2021)]. The code is free to use for academic purpose only.

Link:

• Shuang Ma, Zhaoyang Zeng, Daniel McDuff, Yale Song. Contrastive Learning of Global and Local Audio-Visual Representations.
• [Arxiv]
• [OpenReview]

Pretrain Instruction

Global-local pretraining on Kinetics

create conda environment:

conda env create -f conda_env.yml

Prepare dataset:

• Kinetics: [Download]
• UCF101: [Download]
• HMDB51: [Download]
• LRW: [Download]

Preprain your own model

bash run.sh