A suitable conda environment named noisy_gait can be created and activated with:
conda env create -f noisy_gait.yaml
conda activate noisy_gait
First please download the original data.
CASIA-B http://www.cbsr.ia.ac.cn/GaitDatasetB-silh.zip
OUMVLP http://www.am.sanken.osaka-u.ac.jp/BiometricDB/GaitMVLP.html
Outdoor-Gait https://drive.google.com/drive/folders/1XRWq40G3Zk03YaELywxuVKNodul4TziG
To create a noisy labels gait dataset on CL condition, use the following operation
python add_noisy_labels.py --src_dir CASIA-B --des_dir Noisy-CL-CASIA-B --noise_rate your_noise_rate
To create a noisy appearance gait dataset, use the following operation, choose the appearance typr as dilate or erode. The default probability is 0.5, the default position is [8,56]
python add_noisy_appearances.py --src_dir CASIA-B --des_dir Noisy-CL-CASIA-B --noise_rate your_noise_rate --clothdilate True --clotherode False
To create a random label noise gait dataset, use the following operation
python add_noisy_randm.py --src_dir CASIA-B --des_dir Noisy-CL-CASIA-B --noise_rate your_noise_rate
You can choose the model zoo e.g. GaitSet, GaitGL in the ./config/XXX.yaml and uncomment the corresponding command in train.sh. Then,
train the model by
sh train.sh
test the model by
sh test.sh
We appreciate the wonderful base implementation from Fan etal. and their effort for keeping updates of this code base.