Privacy-Preserving MAE Align (PPMA)

This repository contains instructions needed to recreate the benchmark detailed in the paper Learning Human Action Recognition Representations Without Real Humans and the code to reproduce the results.

Fig: Illustration of PPMA. Model first pretrained on No-Human Kinetics with MAE followed by joint label-alignment pretraining on Synthetic data and No-Human Kinetics.

Refer to data/ for how to access the benchmark datasets.

Refer to code/ for how to train models with PPMA and evaluate them on this benchmark.