Skip to content

"Can Human Sex Be Learned Using Only 2D Keypoint Estimations?"

License

Notifications You must be signed in to change notification settings

kristijanbartol/human-sex-classifier

Repository files navigation

Human sex classifier

Introduction

An implementation for the "Can Human Sex Be Learned Using Only 2D Keypoint Estimations?" paper (arXiv).

Installation

Use pip3 install requirements.txt or docker build -t sex-recognition ..

Prepare, train and evaluate

To run the training, first download (PETA and/or 3DPeople) and prepare the datasets:

python3 src/prepare_datasets.py --name peta --dataset peta

Then you can run the training:

python3 main.py --name peta --train_datasets peta --test_dataset peta --arch fcn

To evaluate the model, run the experiments multiple times (as input data is small and the architecture is simple, it should take only few minutes per experiment):

./eval peta peta

To get the boxplots and the correlations from the paper, use the scripts from report/ directory:

python3 correlation.py
python3 report.py peta

You can also combine multiple training datasets, for example:

python3 main.py --name peta --train_datasets 3dpeople,peta --test_dataset peta --arch fcn

See more data preparation and training options by:

python3 src/prepare_datasets.py -h
python3 main.py -h

License

MIT

About

"Can Human Sex Be Learned Using Only 2D Keypoint Estimations?"

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published