zihaomu/resgait

ReSGait is a research project focused on recognizing people by how they walk. The repository provides the code and instructions to run benchmark experiments on a dataset of real-world walking videos, so researchers can test how well computer vision systems identify individuals based on their gait. The repository walks you through downloading the dataset, organizing the data labels, and running the experiments. You adjust a configuration file with your training parameters, run a training script, and then run a testing script to evaluate the results. The experiments are based on existing open-source code, including a method called GaitSet. Researchers working on biometric identification would use this to evaluate gait recognition algorithms in real-world scenes. For example, a team developing security or surveillance technology could test whether their system can correctly identify a person walking through a camera's view, accounting for real-world variables like different clothing, carrying bags, or phone use. The README doesn't go into much detail about the dataset itself beyond noting that it includes real-scene walking footage with labels for clothing, activity, gender, carrying items, walking route, subject identity, and date. Notably, the benchmark based on GaitSet is listed as not yet finished, so part of the project's experiments appear to still be in progress.