In this repository I will show how to train and test on IJB-A dataset.
Average Pooling (AVE) face features within a template as the Baseline method.
AVE: TAR=[39.78, 64.20, 84.14, 96.04 ]@FAR=[1e-4, 1e-3, 1e-2, 1e-1]
NAN: TAR=[54.51, 74.54, 87.66, 96.31]@FAR=[1e-4, 1e-3, 1e-2, 1e-1]
-
Before to train the model, please download face features from [BaiduYun] (code: 7zyg) and put the unzipped data into ./data/IJBA/resnet34. The features are extracted by ResNet34 model trained on WebFace dataset with only Softmax Loss.
-
run ./prepare_data/dataset_IJBA.py to prepare the train and test data for IJB-A dataset.
-
run ./train_ijba.py to train and test on IJB-A dataset
You can find the train log at [log] and [eval_result]
If you find any bug, please be free to contact me. My email is yirong.maoATvipl.ict.ac.cn